Datasets:

Modalities:
Text
ArXiv:
Maple222 commited on
Commit
3fa082a
·
verified ·
1 Parent(s): f9dff7a

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. out/eval/openllama_arxiv_mc/2407/config.json +22 -0
  2. out/eval/openllama_arxiv_mc/2407/generation_config.json +7 -0
  3. out/eval/openllama_arxiv_mc/2407/log.txt +63 -0
  4. out/eval/openllama_arxiv_mc/2407/model_config.yaml +44 -0
  5. out/eval/openllama_arxiv_mc/2407/results.json +0 -0
  6. out/eval/openllama_arxiv_mc/2407/tokenizer_config.json +1 -0
  7. out/eval/openllama_arxiv_mc/2407/values.json +286 -0
  8. out/eval/openllama_arxiv_mc/2408/config.json +22 -0
  9. out/eval/openllama_arxiv_mc/2408/generation_config.json +7 -0
  10. out/eval/openllama_arxiv_mc/2408/log.txt +65 -0
  11. out/eval/openllama_arxiv_mc/2408/model_config.yaml +44 -0
  12. out/eval/openllama_arxiv_mc/2408/results.json +0 -0
  13. out/eval/openllama_arxiv_mc/2408/tokenizer_config.json +1 -0
  14. out/eval/openllama_arxiv_mc/2408/values.json +286 -0
  15. out/eval/openllama_arxiv_mc/2409/config.json +22 -0
  16. out/eval/openllama_arxiv_mc/2409/generation_config.json +7 -0
  17. out/eval/openllama_arxiv_mc/2409/log.txt +65 -0
  18. out/eval/openllama_arxiv_mc/2409/model_config.yaml +44 -0
  19. out/eval/openllama_arxiv_mc/2409/results.json +0 -0
  20. out/eval/openllama_arxiv_mc/2409/tokenizer_config.json +1 -0
  21. out/eval/openllama_arxiv_mc/2409/values.json +286 -0
  22. out/eval/openllama_arxiv_mc/2410/config.json +22 -0
  23. out/eval/openllama_arxiv_mc/2410/generation_config.json +7 -0
  24. out/eval/openllama_arxiv_mc/2410/log.txt +65 -0
  25. out/eval/openllama_arxiv_mc/2410/model_config.yaml +44 -0
  26. out/eval/openllama_arxiv_mc/2410/results.json +0 -0
  27. out/eval/openllama_arxiv_mc/2410/tokenizer_config.json +1 -0
  28. out/eval/openllama_arxiv_mc/2410/values.json +286 -0
  29. out/eval/openllama_arxiv_mc/2411/config.json +22 -0
  30. out/eval/openllama_arxiv_mc/2411/generation_config.json +7 -0
  31. out/eval/openllama_arxiv_mc/2411/log.txt +65 -0
  32. out/eval/openllama_arxiv_mc/2411/model_config.yaml +44 -0
  33. out/eval/openllama_arxiv_mc/2411/results.json +0 -0
  34. out/eval/openllama_arxiv_mc/2411/tokenizer_config.json +1 -0
  35. out/eval/openllama_arxiv_mc/2411/values.json +286 -0
  36. out/eval/openllama_arxiv_mc/2412/config.json +22 -0
  37. out/eval/openllama_arxiv_mc/2412/generation_config.json +7 -0
  38. out/eval/openllama_arxiv_mc/2412/log.txt +65 -0
  39. out/eval/openllama_arxiv_mc/2412/model_config.yaml +44 -0
  40. out/eval/openllama_arxiv_mc/2412/results.json +0 -0
  41. out/eval/openllama_arxiv_mc/2412/tokenizer_config.json +1 -0
  42. out/eval/openllama_arxiv_mc/2412/values.json +286 -0
  43. out/eval/openllama_arxiv_mc/2501/config.json +22 -0
  44. out/eval/openllama_arxiv_mc/2501/generation_config.json +7 -0
  45. out/eval/openllama_arxiv_mc/2501/log.txt +65 -0
  46. out/eval/openllama_arxiv_mc/2501/model_config.yaml +44 -0
  47. out/eval/openllama_arxiv_mc/2501/tokenizer_config.json +1 -0
  48. out/eval/openllama_benches/2407/config.json +22 -0
  49. out/eval/openllama_benches/2407/generation_config.json +7 -0
  50. out/eval/openllama_benches/2407/log.txt +0 -0
out/eval/openllama_arxiv_mc/2407/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_arxiv_mc/2407/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_arxiv_mc/2407/log.txt ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/100 [00:00<?, ?it/s]
 
1
  0%| | 0/100 [00:00<?, ?it/s]
 
2
  0%| | 0/100 [00:00<?, ?it/s]
 
3
  0%| | 0/100 [00:00<?, ?it/s]
 
4
  0%| | 0/100 [00:00<?, ?it/s]
 
5
  0%| | 0/100 [00:00<?, ?it/s]
 
6
  0%| | 0/100 [00:00<?, ?it/s]
 
7
  0%| | 0/100 [00:00<?, ?it/s]
 
8
  0%| | 0/100 [00:00<?, ?it/s]
 
9
  0%| | 0/100 [00:00<?, ?it/s]
 
10
  0%| | 0/100 [00:00<?, ?it/s]
 
11
  0%| | 0/100 [00:00<?, ?it/s]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'access_token': None,
2
+ 'batch_size': 4,
3
+ 'checkpoint_dir': PosixPath('out/pretrain/openllama/2407/final'),
4
+ 'device': None,
5
+ 'dtype': None,
6
+ 'force_conversion': False,
7
+ 'limit': None,
8
+ 'num_fewshot': None,
9
+ 'out_dir': PosixPath('out/eval/openllama_arxiv_mc/2407'),
10
+ 'save_filepath': None,
11
+ 'seed': 1234,
12
+ 'tasks': 'arxiv_mc'}
13
+ INFO 09-24 14:58:29 [__init__.py:241] Automatically detected platform cuda.
14
+ Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
15
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message
16
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama_fast.LlamaTokenizerFast'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message.
17
+
18
  0%| | 0/100 [00:00<?, ?it/s]
19
+
20
  0%| | 0/100 [00:00<?, ?it/s]
21
+
22
  0%| | 0/100 [00:00<?, ?it/s]
23
+
24
  0%| | 0/100 [00:00<?, ?it/s]
25
+
26
  0%| | 0/100 [00:00<?, ?it/s]
27
+
28
  0%| | 0/100 [00:00<?, ?it/s]
29
+
30
  0%| | 0/100 [00:00<?, ?it/s]
31
+
32
  0%| | 0/100 [00:00<?, ?it/s]
33
+
34
  0%| | 0/100 [00:00<?, ?it/s]
35
+
36
  0%| | 0/100 [00:00<?, ?it/s]
37
+
38
  0%| | 0/100 [00:00<?, ?it/s]
39
+
40
  0%| | 0/100 [00:00<?, ?it/s]
41
+
42
+ | Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
43
+ |----------|-------|------|-----:|--------|---|-----:|---|-----:|
44
+ |arxiv_mc | |none | |acc |↑ |0.1767|± |0.0110|
45
+ | | |none | |acc_norm|↑ |0.2467|± |0.0124|
46
+ | - 2024-07|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
47
+ | | |none | 0|acc_norm|↑ |0.3000|± |0.0461|
48
+ | - 2024-08|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
49
+ | | |none | 0|acc_norm|↑ |0.2100|± |0.0409|
50
+ | - 2024-09|Yaml |none | 0|acc |↑ |0.1500|± |0.0359|
51
+ | | |none | 0|acc_norm|↑ |0.2300|± |0.0423|
52
+ | - 2024-10|Yaml |none | 0|acc |↑ |0.1500|± |0.0359|
53
+ | | |none | 0|acc_norm|↑ |0.2000|± |0.0402|
54
+ | - 2024-11|Yaml |none | 0|acc |↑ |0.2400|± |0.0429|
55
+ | | |none | 0|acc_norm|↑ |0.3500|± |0.0479|
56
+ | - 2024-12|Yaml |none | 0|acc |↑ |0.1300|± |0.0338|
57
+ | | |none | 0|acc_norm|↑ |0.2400|± |0.0429|
58
+ | - 2025-01|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
59
+ | | |none | 0|acc_norm|↑ |0.2500|± |0.0435|
60
+ | - 2025-02|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
61
+ | | |none | 0|acc_norm|↑ |0.2800|± |0.0451|
62
+ | - 2025-03|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
63
+ | | |none | 0|acc_norm|↑ |0.2200|± |0.0416|
64
+ | - 2025-04|Yaml |none | 0|acc |↑ |0.1200|± |0.0327|
65
+ | | |none | 0|acc_norm|↑ |0.1500|± |0.0359|
66
+ | - 2025-05|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
67
+ | | |none | 0|acc_norm|↑ |0.2200|± |0.0416|
68
+ | - 2025-06|Yaml |none | 0|acc |↑ |0.1900|± |0.0394|
69
+ | | |none | 0|acc_norm|↑ |0.3100|± |0.0465|
70
+
71
+ | Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
72
+ |--------|-------|------|------|--------|---|-----:|---|-----:|
73
+ |arxiv_mc| |none | |acc |↑ |0.1767|± |0.0110|
74
+ | | |none | |acc_norm|↑ |0.2467|± |0.0124|
75
+
out/eval/openllama_arxiv_mc/2407/model_config.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ attention_logit_softcapping: null
2
+ attention_scores_scalar: null
3
+ attn_bias: false
4
+ bias: false
5
+ block_size: 2048
6
+ final_logit_softcapping: null
7
+ gelu_approximate: none
8
+ head_size: 100
9
+ hf_config:
10
+ name: open_llama_3b
11
+ org: openlm-research
12
+ intermediate_size: 8640
13
+ lm_head_bias: false
14
+ mlp_class_name: LLaMAMLP
15
+ moe_intermediate_size: null
16
+ n_embd: 3200
17
+ n_expert: 0
18
+ n_expert_per_token: 0
19
+ n_head: 32
20
+ n_layer: 26
21
+ n_query_groups: 32
22
+ name: open_llama_3b
23
+ norm_1: true
24
+ norm_2: true
25
+ norm_class_name: RMSNorm
26
+ norm_eps: 1.0e-06
27
+ norm_qk: false
28
+ norm_qk_type: default
29
+ padded_vocab_size: 32000
30
+ padding_multiple: 64
31
+ parallel_residual: false
32
+ post_attention_norm: false
33
+ post_mlp_norm: false
34
+ rope_adjustments: null
35
+ rope_base: 10000
36
+ rope_condense_ratio: 1
37
+ rope_indices: null
38
+ rope_local_base_freq: null
39
+ rotary_percentage: 1.0
40
+ scale_embeddings: false
41
+ shared_attention_norm: false
42
+ sliding_window_indices: null
43
+ sliding_window_size: null
44
+ vocab_size: 32000
out/eval/openllama_arxiv_mc/2407/results.json ADDED
The diff for this file is too large to render. See raw diff
 
out/eval/openllama_arxiv_mc/2407/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"add_bos_token": true, "add_eos_token": false, "model_max_length": 2048, "pad_token": null, "sp_model_kwargs": {}, "tokenizer_class": "LlamaTokenizer", "clean_up_tokenization_spaces": false, "bos_token": {"__type": "AddedToken", "content": "<s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "eos_token": {"__type": "AddedToken", "content": "</s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "unk_token": {"__type": "AddedToken", "content": "<unk>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}}
out/eval/openllama_arxiv_mc/2407/values.json ADDED
@@ -0,0 +1,286 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tasks": [
3
+ {
4
+ "task": "arxiv_mc",
5
+ "metric": "acc_norm",
6
+ "value": 0.24666666666666667
7
+ },
8
+ {
9
+ "task": "arxiv_mc",
10
+ "metric": "acc_norm_stderr",
11
+ "value": 0.012412996881595178
12
+ },
13
+ {
14
+ "task": "arxiv_mc",
15
+ "metric": "acc",
16
+ "value": 0.17666666666666667
17
+ },
18
+ {
19
+ "task": "arxiv_mc",
20
+ "metric": "acc_stderr",
21
+ "value": 0.011023009299040534
22
+ },
23
+ {
24
+ "task": "arxiv_mc_2407",
25
+ "metric": "acc",
26
+ "value": 0.21
27
+ },
28
+ {
29
+ "task": "arxiv_mc_2407",
30
+ "metric": "acc_stderr",
31
+ "value": 0.040936018074033256
32
+ },
33
+ {
34
+ "task": "arxiv_mc_2407",
35
+ "metric": "acc_norm",
36
+ "value": 0.3
37
+ },
38
+ {
39
+ "task": "arxiv_mc_2407",
40
+ "metric": "acc_norm_stderr",
41
+ "value": 0.046056618647183814
42
+ },
43
+ {
44
+ "task": "arxiv_mc_2408",
45
+ "metric": "acc",
46
+ "value": 0.18
47
+ },
48
+ {
49
+ "task": "arxiv_mc_2408",
50
+ "metric": "acc_stderr",
51
+ "value": 0.038612291966536955
52
+ },
53
+ {
54
+ "task": "arxiv_mc_2408",
55
+ "metric": "acc_norm",
56
+ "value": 0.21
57
+ },
58
+ {
59
+ "task": "arxiv_mc_2408",
60
+ "metric": "acc_norm_stderr",
61
+ "value": 0.040936018074033256
62
+ },
63
+ {
64
+ "task": "arxiv_mc_2409",
65
+ "metric": "acc",
66
+ "value": 0.15
67
+ },
68
+ {
69
+ "task": "arxiv_mc_2409",
70
+ "metric": "acc_stderr",
71
+ "value": 0.03588702812826373
72
+ },
73
+ {
74
+ "task": "arxiv_mc_2409",
75
+ "metric": "acc_norm",
76
+ "value": 0.23
77
+ },
78
+ {
79
+ "task": "arxiv_mc_2409",
80
+ "metric": "acc_norm_stderr",
81
+ "value": 0.042295258468165065
82
+ },
83
+ {
84
+ "task": "arxiv_mc_2410",
85
+ "metric": "acc",
86
+ "value": 0.15
87
+ },
88
+ {
89
+ "task": "arxiv_mc_2410",
90
+ "metric": "acc_stderr",
91
+ "value": 0.035887028128263714
92
+ },
93
+ {
94
+ "task": "arxiv_mc_2410",
95
+ "metric": "acc_norm",
96
+ "value": 0.2
97
+ },
98
+ {
99
+ "task": "arxiv_mc_2410",
100
+ "metric": "acc_norm_stderr",
101
+ "value": 0.04020151261036845
102
+ },
103
+ {
104
+ "task": "arxiv_mc_2411",
105
+ "metric": "acc",
106
+ "value": 0.24
107
+ },
108
+ {
109
+ "task": "arxiv_mc_2411",
110
+ "metric": "acc_stderr",
111
+ "value": 0.04292346959909283
112
+ },
113
+ {
114
+ "task": "arxiv_mc_2411",
115
+ "metric": "acc_norm",
116
+ "value": 0.35
117
+ },
118
+ {
119
+ "task": "arxiv_mc_2411",
120
+ "metric": "acc_norm_stderr",
121
+ "value": 0.04793724854411019
122
+ },
123
+ {
124
+ "task": "arxiv_mc_2412",
125
+ "metric": "acc",
126
+ "value": 0.13
127
+ },
128
+ {
129
+ "task": "arxiv_mc_2412",
130
+ "metric": "acc_stderr",
131
+ "value": 0.03379976689896309
132
+ },
133
+ {
134
+ "task": "arxiv_mc_2412",
135
+ "metric": "acc_norm",
136
+ "value": 0.24
137
+ },
138
+ {
139
+ "task": "arxiv_mc_2412",
140
+ "metric": "acc_norm_stderr",
141
+ "value": 0.042923469599092816
142
+ },
143
+ {
144
+ "task": "arxiv_mc_2501",
145
+ "metric": "acc",
146
+ "value": 0.2
147
+ },
148
+ {
149
+ "task": "arxiv_mc_2501",
150
+ "metric": "acc_stderr",
151
+ "value": 0.04020151261036846
152
+ },
153
+ {
154
+ "task": "arxiv_mc_2501",
155
+ "metric": "acc_norm",
156
+ "value": 0.25
157
+ },
158
+ {
159
+ "task": "arxiv_mc_2501",
160
+ "metric": "acc_norm_stderr",
161
+ "value": 0.04351941398892446
162
+ },
163
+ {
164
+ "task": "arxiv_mc_2502",
165
+ "metric": "acc",
166
+ "value": 0.2
167
+ },
168
+ {
169
+ "task": "arxiv_mc_2502",
170
+ "metric": "acc_stderr",
171
+ "value": 0.040201512610368445
172
+ },
173
+ {
174
+ "task": "arxiv_mc_2502",
175
+ "metric": "acc_norm",
176
+ "value": 0.28
177
+ },
178
+ {
179
+ "task": "arxiv_mc_2502",
180
+ "metric": "acc_norm_stderr",
181
+ "value": 0.04512608598542127
182
+ },
183
+ {
184
+ "task": "arxiv_mc_2503",
185
+ "metric": "acc",
186
+ "value": 0.17
187
+ },
188
+ {
189
+ "task": "arxiv_mc_2503",
190
+ "metric": "acc_stderr",
191
+ "value": 0.0377525168068637
192
+ },
193
+ {
194
+ "task": "arxiv_mc_2503",
195
+ "metric": "acc_norm",
196
+ "value": 0.22
197
+ },
198
+ {
199
+ "task": "arxiv_mc_2503",
200
+ "metric": "acc_norm_stderr",
201
+ "value": 0.041633319989322674
202
+ },
203
+ {
204
+ "task": "arxiv_mc_2504",
205
+ "metric": "acc",
206
+ "value": 0.12
207
+ },
208
+ {
209
+ "task": "arxiv_mc_2504",
210
+ "metric": "acc_stderr",
211
+ "value": 0.032659863237109066
212
+ },
213
+ {
214
+ "task": "arxiv_mc_2504",
215
+ "metric": "acc_norm",
216
+ "value": 0.15
217
+ },
218
+ {
219
+ "task": "arxiv_mc_2504",
220
+ "metric": "acc_norm_stderr",
221
+ "value": 0.03588702812826373
222
+ },
223
+ {
224
+ "task": "arxiv_mc_2505",
225
+ "metric": "acc",
226
+ "value": 0.18
227
+ },
228
+ {
229
+ "task": "arxiv_mc_2505",
230
+ "metric": "acc_stderr",
231
+ "value": 0.038612291966536955
232
+ },
233
+ {
234
+ "task": "arxiv_mc_2505",
235
+ "metric": "acc_norm",
236
+ "value": 0.22
237
+ },
238
+ {
239
+ "task": "arxiv_mc_2505",
240
+ "metric": "acc_norm_stderr",
241
+ "value": 0.041633319989322674
242
+ },
243
+ {
244
+ "task": "arxiv_mc_2506",
245
+ "metric": "acc",
246
+ "value": 0.19
247
+ },
248
+ {
249
+ "task": "arxiv_mc_2506",
250
+ "metric": "acc_stderr",
251
+ "value": 0.03942772444036622
252
+ },
253
+ {
254
+ "task": "arxiv_mc_2506",
255
+ "metric": "acc_norm",
256
+ "value": 0.31
257
+ },
258
+ {
259
+ "task": "arxiv_mc_2506",
260
+ "metric": "acc_norm_stderr",
261
+ "value": 0.04648231987117316
262
+ }
263
+ ],
264
+ "groups": [
265
+ {
266
+ "group": "arxiv_mc",
267
+ "metric": "acc_norm",
268
+ "value": 0.24666666666666667
269
+ },
270
+ {
271
+ "group": "arxiv_mc",
272
+ "metric": "acc_norm_stderr",
273
+ "value": 0.012412996881595178
274
+ },
275
+ {
276
+ "group": "arxiv_mc",
277
+ "metric": "acc",
278
+ "value": 0.17666666666666667
279
+ },
280
+ {
281
+ "group": "arxiv_mc",
282
+ "metric": "acc_stderr",
283
+ "value": 0.011023009299040534
284
+ }
285
+ ]
286
+ }
out/eval/openllama_arxiv_mc/2408/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_arxiv_mc/2408/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_arxiv_mc/2408/log.txt ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/100 [00:00<?, ?it/s]
 
1
  0%| | 0/100 [00:00<?, ?it/s]
 
2
  0%| | 0/100 [00:00<?, ?it/s]
 
3
  0%| | 0/100 [00:00<?, ?it/s]
 
4
  0%| | 0/100 [00:00<?, ?it/s]
 
5
  0%| | 0/100 [00:00<?, ?it/s]
 
6
  0%| | 0/100 [00:00<?, ?it/s]
 
7
  0%| | 0/100 [00:00<?, ?it/s]
 
8
  0%| | 0/100 [00:00<?, ?it/s]
 
9
  0%| | 0/100 [00:00<?, ?it/s]
 
10
  0%| | 0/100 [00:00<?, ?it/s]
 
11
  0%| | 0/100 [00:00<?, ?it/s]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'access_token': None,
2
+ 'batch_size': 4,
3
+ 'checkpoint_dir': PosixPath('out/pretrain/openllama/2408/final'),
4
+ 'device': None,
5
+ 'dtype': None,
6
+ 'force_conversion': False,
7
+ 'limit': None,
8
+ 'num_fewshot': None,
9
+ 'out_dir': PosixPath('out/eval/openllama_arxiv_mc/2408'),
10
+ 'save_filepath': None,
11
+ 'seed': 1234,
12
+ 'tasks': 'arxiv_mc'}
13
+ INFO 09-24 14:59:27 [__init__.py:241] Automatically detected platform cuda.
14
+ Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
15
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message
16
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama_fast.LlamaTokenizerFast'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message.
17
+ {'checkpoint_dir': PosixPath('out/pretrain/openllama/2408/final'),
18
+ 'output_dir': PosixPath('out/eval/openllama_arxiv_mc/2408')}
19
+
20
  0%| | 0/100 [00:00<?, ?it/s]
21
+
22
  0%| | 0/100 [00:00<?, ?it/s]
23
+
24
  0%| | 0/100 [00:00<?, ?it/s]
25
+
26
  0%| | 0/100 [00:00<?, ?it/s]
27
+
28
  0%| | 0/100 [00:00<?, ?it/s]
29
+
30
  0%| | 0/100 [00:00<?, ?it/s]
31
+
32
  0%| | 0/100 [00:00<?, ?it/s]
33
+
34
  0%| | 0/100 [00:00<?, ?it/s]
35
+
36
  0%| | 0/100 [00:00<?, ?it/s]
37
+
38
  0%| | 0/100 [00:00<?, ?it/s]
39
+
40
  0%| | 0/100 [00:00<?, ?it/s]
41
+
42
  0%| | 0/100 [00:00<?, ?it/s]
43
+
44
+ | Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
45
+ |----------|-------|------|-----:|--------|---|-----:|---|-----:|
46
+ |arxiv_mc | |none | |acc |↑ |0.1717|± |0.0109|
47
+ | | |none | |acc_norm|↑ |0.2500|± |0.0124|
48
+ | - 2024-07|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
49
+ | | |none | 0|acc_norm|↑ |0.3000|± |0.0461|
50
+ | - 2024-08|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
51
+ | | |none | 0|acc_norm|↑ |0.2000|± |0.0402|
52
+ | - 2024-09|Yaml |none | 0|acc |↑ |0.1600|± |0.0368|
53
+ | | |none | 0|acc_norm|↑ |0.2500|± |0.0435|
54
+ | - 2024-10|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
55
+ | | |none | 0|acc_norm|↑ |0.2200|± |0.0416|
56
+ | - 2024-11|Yaml |none | 0|acc |↑ |0.2500|± |0.0435|
57
+ | | |none | 0|acc_norm|↑ |0.3800|± |0.0488|
58
+ | - 2024-12|Yaml |none | 0|acc |↑ |0.1000|± |0.0302|
59
+ | | |none | 0|acc_norm|↑ |0.2300|± |0.0423|
60
+ | - 2025-01|Yaml |none | 0|acc |↑ |0.1600|± |0.0368|
61
+ | | |none | 0|acc_norm|↑ |0.2400|± |0.0429|
62
+ | - 2025-02|Yaml |none | 0|acc |↑ |0.2200|± |0.0416|
63
+ | | |none | 0|acc_norm|↑ |0.3200|± |0.0469|
64
+ | - 2025-03|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
65
+ | | |none | 0|acc_norm|↑ |0.2200|± |0.0416|
66
+ | - 2025-04|Yaml |none | 0|acc |↑ |0.1200|± |0.0327|
67
+ | | |none | 0|acc_norm|↑ |0.1500|± |0.0359|
68
+ | - 2025-05|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
69
+ | | |none | 0|acc_norm|↑ |0.2000|± |0.0402|
70
+ | - 2025-06|Yaml |none | 0|acc |↑ |0.1500|± |0.0359|
71
+ | | |none | 0|acc_norm|↑ |0.2900|± |0.0456|
72
+
73
+ | Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
74
+ |--------|-------|------|------|--------|---|-----:|---|-----:|
75
+ |arxiv_mc| |none | |acc |↑ |0.1717|± |0.0109|
76
+ | | |none | |acc_norm|↑ |0.2500|± |0.0124|
77
+
out/eval/openllama_arxiv_mc/2408/model_config.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ attention_logit_softcapping: null
2
+ attention_scores_scalar: null
3
+ attn_bias: false
4
+ bias: false
5
+ block_size: 2048
6
+ final_logit_softcapping: null
7
+ gelu_approximate: none
8
+ head_size: 100
9
+ hf_config:
10
+ name: open_llama_3b
11
+ org: openlm-research
12
+ intermediate_size: 8640
13
+ lm_head_bias: false
14
+ mlp_class_name: LLaMAMLP
15
+ moe_intermediate_size: null
16
+ n_embd: 3200
17
+ n_expert: 0
18
+ n_expert_per_token: 0
19
+ n_head: 32
20
+ n_layer: 26
21
+ n_query_groups: 32
22
+ name: open_llama_3b
23
+ norm_1: true
24
+ norm_2: true
25
+ norm_class_name: RMSNorm
26
+ norm_eps: 1.0e-06
27
+ norm_qk: false
28
+ norm_qk_type: default
29
+ padded_vocab_size: 32000
30
+ padding_multiple: 64
31
+ parallel_residual: false
32
+ post_attention_norm: false
33
+ post_mlp_norm: false
34
+ rope_adjustments: null
35
+ rope_base: 10000
36
+ rope_condense_ratio: 1
37
+ rope_indices: null
38
+ rope_local_base_freq: null
39
+ rotary_percentage: 1.0
40
+ scale_embeddings: false
41
+ shared_attention_norm: false
42
+ sliding_window_indices: null
43
+ sliding_window_size: null
44
+ vocab_size: 32000
out/eval/openllama_arxiv_mc/2408/results.json ADDED
The diff for this file is too large to render. See raw diff
 
out/eval/openllama_arxiv_mc/2408/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"add_bos_token": true, "add_eos_token": false, "model_max_length": 2048, "pad_token": null, "sp_model_kwargs": {}, "tokenizer_class": "LlamaTokenizer", "clean_up_tokenization_spaces": false, "bos_token": {"__type": "AddedToken", "content": "<s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "eos_token": {"__type": "AddedToken", "content": "</s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "unk_token": {"__type": "AddedToken", "content": "<unk>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}}
out/eval/openllama_arxiv_mc/2408/values.json ADDED
@@ -0,0 +1,286 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tasks": [
3
+ {
4
+ "task": "arxiv_mc",
5
+ "metric": "acc_norm",
6
+ "value": 0.25
7
+ },
8
+ {
9
+ "task": "arxiv_mc",
10
+ "metric": "acc_norm_stderr",
11
+ "value": 0.012441783626071456
12
+ },
13
+ {
14
+ "task": "arxiv_mc",
15
+ "metric": "acc",
16
+ "value": 0.17166666666666666
17
+ },
18
+ {
19
+ "task": "arxiv_mc",
20
+ "metric": "acc_stderr",
21
+ "value": 0.010880820552341446
22
+ },
23
+ {
24
+ "task": "arxiv_mc_2407",
25
+ "metric": "acc",
26
+ "value": 0.21
27
+ },
28
+ {
29
+ "task": "arxiv_mc_2407",
30
+ "metric": "acc_stderr",
31
+ "value": 0.040936018074033256
32
+ },
33
+ {
34
+ "task": "arxiv_mc_2407",
35
+ "metric": "acc_norm",
36
+ "value": 0.3
37
+ },
38
+ {
39
+ "task": "arxiv_mc_2407",
40
+ "metric": "acc_norm_stderr",
41
+ "value": 0.046056618647183814
42
+ },
43
+ {
44
+ "task": "arxiv_mc_2408",
45
+ "metric": "acc",
46
+ "value": 0.17
47
+ },
48
+ {
49
+ "task": "arxiv_mc_2408",
50
+ "metric": "acc_stderr",
51
+ "value": 0.03775251680686371
52
+ },
53
+ {
54
+ "task": "arxiv_mc_2408",
55
+ "metric": "acc_norm",
56
+ "value": 0.2
57
+ },
58
+ {
59
+ "task": "arxiv_mc_2408",
60
+ "metric": "acc_norm_stderr",
61
+ "value": 0.04020151261036844
62
+ },
63
+ {
64
+ "task": "arxiv_mc_2409",
65
+ "metric": "acc",
66
+ "value": 0.16
67
+ },
68
+ {
69
+ "task": "arxiv_mc_2409",
70
+ "metric": "acc_stderr",
71
+ "value": 0.0368452949177471
72
+ },
73
+ {
74
+ "task": "arxiv_mc_2409",
75
+ "metric": "acc_norm",
76
+ "value": 0.25
77
+ },
78
+ {
79
+ "task": "arxiv_mc_2409",
80
+ "metric": "acc_norm_stderr",
81
+ "value": 0.04351941398892446
82
+ },
83
+ {
84
+ "task": "arxiv_mc_2410",
85
+ "metric": "acc",
86
+ "value": 0.18
87
+ },
88
+ {
89
+ "task": "arxiv_mc_2410",
90
+ "metric": "acc_stderr",
91
+ "value": 0.03861229196653697
92
+ },
93
+ {
94
+ "task": "arxiv_mc_2410",
95
+ "metric": "acc_norm",
96
+ "value": 0.22
97
+ },
98
+ {
99
+ "task": "arxiv_mc_2410",
100
+ "metric": "acc_norm_stderr",
101
+ "value": 0.04163331998932269
102
+ },
103
+ {
104
+ "task": "arxiv_mc_2411",
105
+ "metric": "acc",
106
+ "value": 0.25
107
+ },
108
+ {
109
+ "task": "arxiv_mc_2411",
110
+ "metric": "acc_stderr",
111
+ "value": 0.04351941398892446
112
+ },
113
+ {
114
+ "task": "arxiv_mc_2411",
115
+ "metric": "acc_norm",
116
+ "value": 0.38
117
+ },
118
+ {
119
+ "task": "arxiv_mc_2411",
120
+ "metric": "acc_norm_stderr",
121
+ "value": 0.04878317312145633
122
+ },
123
+ {
124
+ "task": "arxiv_mc_2412",
125
+ "metric": "acc",
126
+ "value": 0.1
127
+ },
128
+ {
129
+ "task": "arxiv_mc_2412",
130
+ "metric": "acc_stderr",
131
+ "value": 0.030151134457776348
132
+ },
133
+ {
134
+ "task": "arxiv_mc_2412",
135
+ "metric": "acc_norm",
136
+ "value": 0.23
137
+ },
138
+ {
139
+ "task": "arxiv_mc_2412",
140
+ "metric": "acc_norm_stderr",
141
+ "value": 0.04229525846816505
142
+ },
143
+ {
144
+ "task": "arxiv_mc_2501",
145
+ "metric": "acc",
146
+ "value": 0.16
147
+ },
148
+ {
149
+ "task": "arxiv_mc_2501",
150
+ "metric": "acc_stderr",
151
+ "value": 0.036845294917747094
152
+ },
153
+ {
154
+ "task": "arxiv_mc_2501",
155
+ "metric": "acc_norm",
156
+ "value": 0.24
157
+ },
158
+ {
159
+ "task": "arxiv_mc_2501",
160
+ "metric": "acc_norm_stderr",
161
+ "value": 0.04292346959909284
162
+ },
163
+ {
164
+ "task": "arxiv_mc_2502",
165
+ "metric": "acc",
166
+ "value": 0.22
167
+ },
168
+ {
169
+ "task": "arxiv_mc_2502",
170
+ "metric": "acc_stderr",
171
+ "value": 0.0416333199893227
172
+ },
173
+ {
174
+ "task": "arxiv_mc_2502",
175
+ "metric": "acc_norm",
176
+ "value": 0.32
177
+ },
178
+ {
179
+ "task": "arxiv_mc_2502",
180
+ "metric": "acc_norm_stderr",
181
+ "value": 0.046882617226215034
182
+ },
183
+ {
184
+ "task": "arxiv_mc_2503",
185
+ "metric": "acc",
186
+ "value": 0.17
187
+ },
188
+ {
189
+ "task": "arxiv_mc_2503",
190
+ "metric": "acc_stderr",
191
+ "value": 0.0377525168068637
192
+ },
193
+ {
194
+ "task": "arxiv_mc_2503",
195
+ "metric": "acc_norm",
196
+ "value": 0.22
197
+ },
198
+ {
199
+ "task": "arxiv_mc_2503",
200
+ "metric": "acc_norm_stderr",
201
+ "value": 0.041633319989322695
202
+ },
203
+ {
204
+ "task": "arxiv_mc_2504",
205
+ "metric": "acc",
206
+ "value": 0.12
207
+ },
208
+ {
209
+ "task": "arxiv_mc_2504",
210
+ "metric": "acc_stderr",
211
+ "value": 0.03265986323710906
212
+ },
213
+ {
214
+ "task": "arxiv_mc_2504",
215
+ "metric": "acc_norm",
216
+ "value": 0.15
217
+ },
218
+ {
219
+ "task": "arxiv_mc_2504",
220
+ "metric": "acc_norm_stderr",
221
+ "value": 0.03588702812826373
222
+ },
223
+ {
224
+ "task": "arxiv_mc_2505",
225
+ "metric": "acc",
226
+ "value": 0.17
227
+ },
228
+ {
229
+ "task": "arxiv_mc_2505",
230
+ "metric": "acc_stderr",
231
+ "value": 0.03775251680686371
232
+ },
233
+ {
234
+ "task": "arxiv_mc_2505",
235
+ "metric": "acc_norm",
236
+ "value": 0.2
237
+ },
238
+ {
239
+ "task": "arxiv_mc_2505",
240
+ "metric": "acc_norm_stderr",
241
+ "value": 0.04020151261036846
242
+ },
243
+ {
244
+ "task": "arxiv_mc_2506",
245
+ "metric": "acc",
246
+ "value": 0.15
247
+ },
248
+ {
249
+ "task": "arxiv_mc_2506",
250
+ "metric": "acc_stderr",
251
+ "value": 0.03588702812826371
252
+ },
253
+ {
254
+ "task": "arxiv_mc_2506",
255
+ "metric": "acc_norm",
256
+ "value": 0.29
257
+ },
258
+ {
259
+ "task": "arxiv_mc_2506",
260
+ "metric": "acc_norm_stderr",
261
+ "value": 0.04560480215720684
262
+ }
263
+ ],
264
+ "groups": [
265
+ {
266
+ "group": "arxiv_mc",
267
+ "metric": "acc_norm",
268
+ "value": 0.25
269
+ },
270
+ {
271
+ "group": "arxiv_mc",
272
+ "metric": "acc_norm_stderr",
273
+ "value": 0.012441783626071456
274
+ },
275
+ {
276
+ "group": "arxiv_mc",
277
+ "metric": "acc",
278
+ "value": 0.17166666666666666
279
+ },
280
+ {
281
+ "group": "arxiv_mc",
282
+ "metric": "acc_stderr",
283
+ "value": 0.010880820552341446
284
+ }
285
+ ]
286
+ }
out/eval/openllama_arxiv_mc/2409/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_arxiv_mc/2409/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_arxiv_mc/2409/log.txt ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/100 [00:00<?, ?it/s]
 
1
  0%| | 0/100 [00:00<?, ?it/s]
 
2
  0%| | 0/100 [00:00<?, ?it/s]
 
3
  0%| | 0/100 [00:00<?, ?it/s]
 
4
  0%| | 0/100 [00:00<?, ?it/s]
 
5
  0%| | 0/100 [00:00<?, ?it/s]
 
6
  0%| | 0/100 [00:00<?, ?it/s]
 
7
  0%| | 0/100 [00:00<?, ?it/s]
 
8
  0%| | 0/100 [00:00<?, ?it/s]
 
9
  0%| | 0/100 [00:00<?, ?it/s]
 
10
  0%| | 0/100 [00:00<?, ?it/s]
 
11
  0%| | 0/100 [00:00<?, ?it/s]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'access_token': None,
2
+ 'batch_size': 4,
3
+ 'checkpoint_dir': PosixPath('out/pretrain/openllama/2409/final'),
4
+ 'device': None,
5
+ 'dtype': None,
6
+ 'force_conversion': False,
7
+ 'limit': None,
8
+ 'num_fewshot': None,
9
+ 'out_dir': PosixPath('out/eval/openllama_arxiv_mc/2409'),
10
+ 'save_filepath': None,
11
+ 'seed': 1234,
12
+ 'tasks': 'arxiv_mc'}
13
+ INFO 09-24 15:01:17 [__init__.py:241] Automatically detected platform cuda.
14
+ Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
15
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message
16
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama_fast.LlamaTokenizerFast'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message.
17
+ {'checkpoint_dir': PosixPath('out/pretrain/openllama/2409/final'),
18
+ 'output_dir': PosixPath('out/eval/openllama_arxiv_mc/2409')}
19
+
20
  0%| | 0/100 [00:00<?, ?it/s]
21
+
22
  0%| | 0/100 [00:00<?, ?it/s]
23
+
24
  0%| | 0/100 [00:00<?, ?it/s]
25
+
26
  0%| | 0/100 [00:00<?, ?it/s]
27
+
28
  0%| | 0/100 [00:00<?, ?it/s]
29
+
30
  0%| | 0/100 [00:00<?, ?it/s]
31
+
32
  0%| | 0/100 [00:00<?, ?it/s]
33
+
34
  0%| | 0/100 [00:00<?, ?it/s]
35
+
36
  0%| | 0/100 [00:00<?, ?it/s]
37
+
38
  0%| | 0/100 [00:00<?, ?it/s]
39
+
40
  0%| | 0/100 [00:00<?, ?it/s]
41
+
42
  0%| | 0/100 [00:00<?, ?it/s]
43
+
44
+ | Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
45
+ |----------|-------|------|-----:|--------|---|-----:|---|-----:|
46
+ |arxiv_mc | |none | |acc |↑ |0.1808|± |0.0111|
47
+ | | |none | |acc_norm|↑ |0.2550|± |0.0125|
48
+ | - 2024-07|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
49
+ | | |none | 0|acc_norm|↑ |0.3200|± |0.0469|
50
+ | - 2024-08|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
51
+ | | |none | 0|acc_norm|↑ |0.2000|± |0.0402|
52
+ | - 2024-09|Yaml |none | 0|acc |↑ |0.1500|± |0.0359|
53
+ | | |none | 0|acc_norm|↑ |0.2500|± |0.0435|
54
+ | - 2024-10|Yaml |none | 0|acc |↑ |0.1600|± |0.0368|
55
+ | | |none | 0|acc_norm|↑ |0.1900|± |0.0394|
56
+ | - 2024-11|Yaml |none | 0|acc |↑ |0.2600|± |0.0441|
57
+ | | |none | 0|acc_norm|↑ |0.3700|± |0.0485|
58
+ | - 2024-12|Yaml |none | 0|acc |↑ |0.1200|± |0.0327|
59
+ | | |none | 0|acc_norm|↑ |0.2300|± |0.0423|
60
+ | - 2025-01|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
61
+ | | |none | 0|acc_norm|↑ |0.2800|± |0.0451|
62
+ | - 2025-02|Yaml |none | 0|acc |↑ |0.1900|± |0.0394|
63
+ | | |none | 0|acc_norm|↑ |0.2800|± |0.0451|
64
+ | - 2025-03|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
65
+ | | |none | 0|acc_norm|↑ |0.2300|± |0.0423|
66
+ | - 2025-04|Yaml |none | 0|acc |↑ |0.1200|± |0.0327|
67
+ | | |none | 0|acc_norm|↑ |0.1800|± |0.0386|
68
+ | - 2025-05|Yaml |none | 0|acc |↑ |0.1600|± |0.0368|
69
+ | | |none | 0|acc_norm|↑ |0.2300|± |0.0423|
70
+ | - 2025-06|Yaml |none | 0|acc |↑ |0.1900|± |0.0394|
71
+ | | |none | 0|acc_norm|↑ |0.3000|± |0.0461|
72
+
73
+ | Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
74
+ |--------|-------|------|------|--------|---|-----:|---|-----:|
75
+ |arxiv_mc| |none | |acc |↑ |0.1808|± |0.0111|
76
+ | | |none | |acc_norm|↑ |0.2550|± |0.0125|
77
+
out/eval/openllama_arxiv_mc/2409/model_config.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ attention_logit_softcapping: null
2
+ attention_scores_scalar: null
3
+ attn_bias: false
4
+ bias: false
5
+ block_size: 2048
6
+ final_logit_softcapping: null
7
+ gelu_approximate: none
8
+ head_size: 100
9
+ hf_config:
10
+ name: open_llama_3b
11
+ org: openlm-research
12
+ intermediate_size: 8640
13
+ lm_head_bias: false
14
+ mlp_class_name: LLaMAMLP
15
+ moe_intermediate_size: null
16
+ n_embd: 3200
17
+ n_expert: 0
18
+ n_expert_per_token: 0
19
+ n_head: 32
20
+ n_layer: 26
21
+ n_query_groups: 32
22
+ name: open_llama_3b
23
+ norm_1: true
24
+ norm_2: true
25
+ norm_class_name: RMSNorm
26
+ norm_eps: 1.0e-06
27
+ norm_qk: false
28
+ norm_qk_type: default
29
+ padded_vocab_size: 32000
30
+ padding_multiple: 64
31
+ parallel_residual: false
32
+ post_attention_norm: false
33
+ post_mlp_norm: false
34
+ rope_adjustments: null
35
+ rope_base: 10000
36
+ rope_condense_ratio: 1
37
+ rope_indices: null
38
+ rope_local_base_freq: null
39
+ rotary_percentage: 1.0
40
+ scale_embeddings: false
41
+ shared_attention_norm: false
42
+ sliding_window_indices: null
43
+ sliding_window_size: null
44
+ vocab_size: 32000
out/eval/openllama_arxiv_mc/2409/results.json ADDED
The diff for this file is too large to render. See raw diff
 
out/eval/openllama_arxiv_mc/2409/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"add_bos_token": true, "add_eos_token": false, "model_max_length": 2048, "pad_token": null, "sp_model_kwargs": {}, "tokenizer_class": "LlamaTokenizer", "clean_up_tokenization_spaces": false, "bos_token": {"__type": "AddedToken", "content": "<s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "eos_token": {"__type": "AddedToken", "content": "</s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "unk_token": {"__type": "AddedToken", "content": "<unk>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}}
out/eval/openllama_arxiv_mc/2409/values.json ADDED
@@ -0,0 +1,286 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tasks": [
3
+ {
4
+ "task": "arxiv_mc",
5
+ "metric": "acc_norm",
6
+ "value": 0.255
7
+ },
8
+ {
9
+ "task": "arxiv_mc",
10
+ "metric": "acc_norm_stderr",
11
+ "value": 0.012546769966334092
12
+ },
13
+ {
14
+ "task": "arxiv_mc",
15
+ "metric": "acc",
16
+ "value": 0.18083333333333335
17
+ },
18
+ {
19
+ "task": "arxiv_mc",
20
+ "metric": "acc_stderr",
21
+ "value": 0.011109532716172954
22
+ },
23
+ {
24
+ "task": "arxiv_mc_2407",
25
+ "metric": "acc",
26
+ "value": 0.2
27
+ },
28
+ {
29
+ "task": "arxiv_mc_2407",
30
+ "metric": "acc_stderr",
31
+ "value": 0.04020151261036846
32
+ },
33
+ {
34
+ "task": "arxiv_mc_2407",
35
+ "metric": "acc_norm",
36
+ "value": 0.32
37
+ },
38
+ {
39
+ "task": "arxiv_mc_2407",
40
+ "metric": "acc_norm_stderr",
41
+ "value": 0.04688261722621504
42
+ },
43
+ {
44
+ "task": "arxiv_mc_2408",
45
+ "metric": "acc",
46
+ "value": 0.21
47
+ },
48
+ {
49
+ "task": "arxiv_mc_2408",
50
+ "metric": "acc_stderr",
51
+ "value": 0.040936018074033256
52
+ },
53
+ {
54
+ "task": "arxiv_mc_2408",
55
+ "metric": "acc_norm",
56
+ "value": 0.2
57
+ },
58
+ {
59
+ "task": "arxiv_mc_2408",
60
+ "metric": "acc_norm_stderr",
61
+ "value": 0.04020151261036843
62
+ },
63
+ {
64
+ "task": "arxiv_mc_2409",
65
+ "metric": "acc",
66
+ "value": 0.15
67
+ },
68
+ {
69
+ "task": "arxiv_mc_2409",
70
+ "metric": "acc_stderr",
71
+ "value": 0.03588702812826373
72
+ },
73
+ {
74
+ "task": "arxiv_mc_2409",
75
+ "metric": "acc_norm",
76
+ "value": 0.25
77
+ },
78
+ {
79
+ "task": "arxiv_mc_2409",
80
+ "metric": "acc_norm_stderr",
81
+ "value": 0.04351941398892446
82
+ },
83
+ {
84
+ "task": "arxiv_mc_2410",
85
+ "metric": "acc",
86
+ "value": 0.16
87
+ },
88
+ {
89
+ "task": "arxiv_mc_2410",
90
+ "metric": "acc_stderr",
91
+ "value": 0.03684529491774709
92
+ },
93
+ {
94
+ "task": "arxiv_mc_2410",
95
+ "metric": "acc_norm",
96
+ "value": 0.19
97
+ },
98
+ {
99
+ "task": "arxiv_mc_2410",
100
+ "metric": "acc_norm_stderr",
101
+ "value": 0.039427724440366234
102
+ },
103
+ {
104
+ "task": "arxiv_mc_2411",
105
+ "metric": "acc",
106
+ "value": 0.26
107
+ },
108
+ {
109
+ "task": "arxiv_mc_2411",
110
+ "metric": "acc_stderr",
111
+ "value": 0.04408440022768078
112
+ },
113
+ {
114
+ "task": "arxiv_mc_2411",
115
+ "metric": "acc_norm",
116
+ "value": 0.37
117
+ },
118
+ {
119
+ "task": "arxiv_mc_2411",
120
+ "metric": "acc_norm_stderr",
121
+ "value": 0.04852365870939099
122
+ },
123
+ {
124
+ "task": "arxiv_mc_2412",
125
+ "metric": "acc",
126
+ "value": 0.12
127
+ },
128
+ {
129
+ "task": "arxiv_mc_2412",
130
+ "metric": "acc_stderr",
131
+ "value": 0.03265986323710906
132
+ },
133
+ {
134
+ "task": "arxiv_mc_2412",
135
+ "metric": "acc_norm",
136
+ "value": 0.23
137
+ },
138
+ {
139
+ "task": "arxiv_mc_2412",
140
+ "metric": "acc_norm_stderr",
141
+ "value": 0.04229525846816505
142
+ },
143
+ {
144
+ "task": "arxiv_mc_2501",
145
+ "metric": "acc",
146
+ "value": 0.21
147
+ },
148
+ {
149
+ "task": "arxiv_mc_2501",
150
+ "metric": "acc_stderr",
151
+ "value": 0.040936018074033256
152
+ },
153
+ {
154
+ "task": "arxiv_mc_2501",
155
+ "metric": "acc_norm",
156
+ "value": 0.28
157
+ },
158
+ {
159
+ "task": "arxiv_mc_2501",
160
+ "metric": "acc_norm_stderr",
161
+ "value": 0.04512608598542127
162
+ },
163
+ {
164
+ "task": "arxiv_mc_2502",
165
+ "metric": "acc",
166
+ "value": 0.19
167
+ },
168
+ {
169
+ "task": "arxiv_mc_2502",
170
+ "metric": "acc_stderr",
171
+ "value": 0.03942772444036623
172
+ },
173
+ {
174
+ "task": "arxiv_mc_2502",
175
+ "metric": "acc_norm",
176
+ "value": 0.28
177
+ },
178
+ {
179
+ "task": "arxiv_mc_2502",
180
+ "metric": "acc_norm_stderr",
181
+ "value": 0.04512608598542127
182
+ },
183
+ {
184
+ "task": "arxiv_mc_2503",
185
+ "metric": "acc",
186
+ "value": 0.2
187
+ },
188
+ {
189
+ "task": "arxiv_mc_2503",
190
+ "metric": "acc_stderr",
191
+ "value": 0.04020151261036846
192
+ },
193
+ {
194
+ "task": "arxiv_mc_2503",
195
+ "metric": "acc_norm",
196
+ "value": 0.23
197
+ },
198
+ {
199
+ "task": "arxiv_mc_2503",
200
+ "metric": "acc_norm_stderr",
201
+ "value": 0.04229525846816506
202
+ },
203
+ {
204
+ "task": "arxiv_mc_2504",
205
+ "metric": "acc",
206
+ "value": 0.12
207
+ },
208
+ {
209
+ "task": "arxiv_mc_2504",
210
+ "metric": "acc_stderr",
211
+ "value": 0.032659863237109066
212
+ },
213
+ {
214
+ "task": "arxiv_mc_2504",
215
+ "metric": "acc_norm",
216
+ "value": 0.18
217
+ },
218
+ {
219
+ "task": "arxiv_mc_2504",
220
+ "metric": "acc_norm_stderr",
221
+ "value": 0.03861229196653697
222
+ },
223
+ {
224
+ "task": "arxiv_mc_2505",
225
+ "metric": "acc",
226
+ "value": 0.16
227
+ },
228
+ {
229
+ "task": "arxiv_mc_2505",
230
+ "metric": "acc_stderr",
231
+ "value": 0.036845294917747094
232
+ },
233
+ {
234
+ "task": "arxiv_mc_2505",
235
+ "metric": "acc_norm",
236
+ "value": 0.23
237
+ },
238
+ {
239
+ "task": "arxiv_mc_2505",
240
+ "metric": "acc_norm_stderr",
241
+ "value": 0.04229525846816507
242
+ },
243
+ {
244
+ "task": "arxiv_mc_2506",
245
+ "metric": "acc",
246
+ "value": 0.19
247
+ },
248
+ {
249
+ "task": "arxiv_mc_2506",
250
+ "metric": "acc_stderr",
251
+ "value": 0.03942772444036623
252
+ },
253
+ {
254
+ "task": "arxiv_mc_2506",
255
+ "metric": "acc_norm",
256
+ "value": 0.3
257
+ },
258
+ {
259
+ "task": "arxiv_mc_2506",
260
+ "metric": "acc_norm_stderr",
261
+ "value": 0.046056618647183814
262
+ }
263
+ ],
264
+ "groups": [
265
+ {
266
+ "group": "arxiv_mc",
267
+ "metric": "acc_norm",
268
+ "value": 0.255
269
+ },
270
+ {
271
+ "group": "arxiv_mc",
272
+ "metric": "acc_norm_stderr",
273
+ "value": 0.012546769966334092
274
+ },
275
+ {
276
+ "group": "arxiv_mc",
277
+ "metric": "acc",
278
+ "value": 0.18083333333333335
279
+ },
280
+ {
281
+ "group": "arxiv_mc",
282
+ "metric": "acc_stderr",
283
+ "value": 0.011109532716172954
284
+ }
285
+ ]
286
+ }
out/eval/openllama_arxiv_mc/2410/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_arxiv_mc/2410/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_arxiv_mc/2410/log.txt ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/100 [00:00<?, ?it/s]
 
1
  0%| | 0/100 [00:00<?, ?it/s]
 
2
  0%| | 0/100 [00:00<?, ?it/s]
 
3
  0%| | 0/100 [00:00<?, ?it/s]
 
4
  0%| | 0/100 [00:00<?, ?it/s]
 
5
  0%| | 0/100 [00:00<?, ?it/s]
 
6
  0%| | 0/100 [00:00<?, ?it/s]
 
7
  0%| | 0/100 [00:00<?, ?it/s]
 
8
  0%| | 0/100 [00:00<?, ?it/s]
 
9
  0%| | 0/100 [00:00<?, ?it/s]
 
10
  0%| | 0/100 [00:00<?, ?it/s]
 
11
  0%| | 0/100 [00:00<?, ?it/s]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'access_token': None,
2
+ 'batch_size': 4,
3
+ 'checkpoint_dir': PosixPath('out/pretrain/openllama/2410/final'),
4
+ 'device': None,
5
+ 'dtype': None,
6
+ 'force_conversion': False,
7
+ 'limit': None,
8
+ 'num_fewshot': None,
9
+ 'out_dir': PosixPath('out/eval/openllama_arxiv_mc/2410'),
10
+ 'save_filepath': None,
11
+ 'seed': 1234,
12
+ 'tasks': 'arxiv_mc'}
13
+ INFO 09-24 15:03:04 [__init__.py:241] Automatically detected platform cuda.
14
+ Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
15
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message
16
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama_fast.LlamaTokenizerFast'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message.
17
+ {'checkpoint_dir': PosixPath('out/pretrain/openllama/2410/final'),
18
+ 'output_dir': PosixPath('out/eval/openllama_arxiv_mc/2410')}
19
+
20
  0%| | 0/100 [00:00<?, ?it/s]
21
+
22
  0%| | 0/100 [00:00<?, ?it/s]
23
+
24
  0%| | 0/100 [00:00<?, ?it/s]
25
+
26
  0%| | 0/100 [00:00<?, ?it/s]
27
+
28
  0%| | 0/100 [00:00<?, ?it/s]
29
+
30
  0%| | 0/100 [00:00<?, ?it/s]
31
+
32
  0%| | 0/100 [00:00<?, ?it/s]
33
+
34
  0%| | 0/100 [00:00<?, ?it/s]
35
+
36
  0%| | 0/100 [00:00<?, ?it/s]
37
+
38
  0%| | 0/100 [00:00<?, ?it/s]
39
+
40
  0%| | 0/100 [00:00<?, ?it/s]
41
+
42
  0%| | 0/100 [00:00<?, ?it/s]
43
+
44
+ | Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
45
+ |----------|-------|------|-----:|--------|---|-----:|---|-----:|
46
+ |arxiv_mc | |none | |acc |↑ |0.1767|± |0.0110|
47
+ | | |none | |acc_norm|↑ |0.2650|± |0.0127|
48
+ | - 2024-07|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
49
+ | | |none | 0|acc_norm|↑ |0.3100|± |0.0465|
50
+ | - 2024-08|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
51
+ | | |none | 0|acc_norm|↑ |0.2100|± |0.0409|
52
+ | - 2024-09|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
53
+ | | |none | 0|acc_norm|↑ |0.3000|± |0.0461|
54
+ | - 2024-10|Yaml |none | 0|acc |↑ |0.1300|± |0.0338|
55
+ | | |none | 0|acc_norm|↑ |0.2100|± |0.0409|
56
+ | - 2024-11|Yaml |none | 0|acc |↑ |0.2600|± |0.0441|
57
+ | | |none | 0|acc_norm|↑ |0.4000|± |0.0492|
58
+ | - 2024-12|Yaml |none | 0|acc |↑ |0.1200|± |0.0327|
59
+ | | |none | 0|acc_norm|↑ |0.2200|± |0.0416|
60
+ | - 2025-01|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
61
+ | | |none | 0|acc_norm|↑ |0.2300|± |0.0423|
62
+ | - 2025-02|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
63
+ | | |none | 0|acc_norm|↑ |0.2600|± |0.0441|
64
+ | - 2025-03|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
65
+ | | |none | 0|acc_norm|↑ |0.2700|± |0.0446|
66
+ | - 2025-04|Yaml |none | 0|acc |↑ |0.1100|± |0.0314|
67
+ | | |none | 0|acc_norm|↑ |0.2000|± |0.0402|
68
+ | - 2025-05|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
69
+ | | |none | 0|acc_norm|↑ |0.2100|± |0.0409|
70
+ | - 2025-06|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
71
+ | | |none | 0|acc_norm|↑ |0.3600|± |0.0482|
72
+
73
+ | Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
74
+ |--------|-------|------|------|--------|---|-----:|---|-----:|
75
+ |arxiv_mc| |none | |acc |↑ |0.1767|± |0.0110|
76
+ | | |none | |acc_norm|↑ |0.2650|± |0.0127|
77
+
out/eval/openllama_arxiv_mc/2410/model_config.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ attention_logit_softcapping: null
2
+ attention_scores_scalar: null
3
+ attn_bias: false
4
+ bias: false
5
+ block_size: 2048
6
+ final_logit_softcapping: null
7
+ gelu_approximate: none
8
+ head_size: 100
9
+ hf_config:
10
+ name: open_llama_3b
11
+ org: openlm-research
12
+ intermediate_size: 8640
13
+ lm_head_bias: false
14
+ mlp_class_name: LLaMAMLP
15
+ moe_intermediate_size: null
16
+ n_embd: 3200
17
+ n_expert: 0
18
+ n_expert_per_token: 0
19
+ n_head: 32
20
+ n_layer: 26
21
+ n_query_groups: 32
22
+ name: open_llama_3b
23
+ norm_1: true
24
+ norm_2: true
25
+ norm_class_name: RMSNorm
26
+ norm_eps: 1.0e-06
27
+ norm_qk: false
28
+ norm_qk_type: default
29
+ padded_vocab_size: 32000
30
+ padding_multiple: 64
31
+ parallel_residual: false
32
+ post_attention_norm: false
33
+ post_mlp_norm: false
34
+ rope_adjustments: null
35
+ rope_base: 10000
36
+ rope_condense_ratio: 1
37
+ rope_indices: null
38
+ rope_local_base_freq: null
39
+ rotary_percentage: 1.0
40
+ scale_embeddings: false
41
+ shared_attention_norm: false
42
+ sliding_window_indices: null
43
+ sliding_window_size: null
44
+ vocab_size: 32000
out/eval/openllama_arxiv_mc/2410/results.json ADDED
The diff for this file is too large to render. See raw diff
 
out/eval/openllama_arxiv_mc/2410/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"add_bos_token": true, "add_eos_token": false, "model_max_length": 2048, "pad_token": null, "sp_model_kwargs": {}, "tokenizer_class": "LlamaTokenizer", "clean_up_tokenization_spaces": false, "bos_token": {"__type": "AddedToken", "content": "<s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "eos_token": {"__type": "AddedToken", "content": "</s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "unk_token": {"__type": "AddedToken", "content": "<unk>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}}
out/eval/openllama_arxiv_mc/2410/values.json ADDED
@@ -0,0 +1,286 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tasks": [
3
+ {
4
+ "task": "arxiv_mc",
5
+ "metric": "acc",
6
+ "value": 0.17666666666666667
7
+ },
8
+ {
9
+ "task": "arxiv_mc",
10
+ "metric": "acc_stderr",
11
+ "value": 0.011004539416375157
12
+ },
13
+ {
14
+ "task": "arxiv_mc",
15
+ "metric": "acc_norm",
16
+ "value": 0.265
17
+ },
18
+ {
19
+ "task": "arxiv_mc",
20
+ "metric": "acc_norm_stderr",
21
+ "value": 0.012674705051722038
22
+ },
23
+ {
24
+ "task": "arxiv_mc_2407",
25
+ "metric": "acc",
26
+ "value": 0.21
27
+ },
28
+ {
29
+ "task": "arxiv_mc_2407",
30
+ "metric": "acc_stderr",
31
+ "value": 0.040936018074033256
32
+ },
33
+ {
34
+ "task": "arxiv_mc_2407",
35
+ "metric": "acc_norm",
36
+ "value": 0.31
37
+ },
38
+ {
39
+ "task": "arxiv_mc_2407",
40
+ "metric": "acc_norm_stderr",
41
+ "value": 0.04648231987117316
42
+ },
43
+ {
44
+ "task": "arxiv_mc_2408",
45
+ "metric": "acc",
46
+ "value": 0.18
47
+ },
48
+ {
49
+ "task": "arxiv_mc_2408",
50
+ "metric": "acc_stderr",
51
+ "value": 0.03861229196653697
52
+ },
53
+ {
54
+ "task": "arxiv_mc_2408",
55
+ "metric": "acc_norm",
56
+ "value": 0.21
57
+ },
58
+ {
59
+ "task": "arxiv_mc_2408",
60
+ "metric": "acc_norm_stderr",
61
+ "value": 0.040936018074033256
62
+ },
63
+ {
64
+ "task": "arxiv_mc_2409",
65
+ "metric": "acc",
66
+ "value": 0.17
67
+ },
68
+ {
69
+ "task": "arxiv_mc_2409",
70
+ "metric": "acc_stderr",
71
+ "value": 0.03775251680686371
72
+ },
73
+ {
74
+ "task": "arxiv_mc_2409",
75
+ "metric": "acc_norm",
76
+ "value": 0.3
77
+ },
78
+ {
79
+ "task": "arxiv_mc_2409",
80
+ "metric": "acc_norm_stderr",
81
+ "value": 0.046056618647183814
82
+ },
83
+ {
84
+ "task": "arxiv_mc_2410",
85
+ "metric": "acc",
86
+ "value": 0.13
87
+ },
88
+ {
89
+ "task": "arxiv_mc_2410",
90
+ "metric": "acc_stderr",
91
+ "value": 0.033799766898963086
92
+ },
93
+ {
94
+ "task": "arxiv_mc_2410",
95
+ "metric": "acc_norm",
96
+ "value": 0.21
97
+ },
98
+ {
99
+ "task": "arxiv_mc_2410",
100
+ "metric": "acc_norm_stderr",
101
+ "value": 0.04093601807403326
102
+ },
103
+ {
104
+ "task": "arxiv_mc_2411",
105
+ "metric": "acc",
106
+ "value": 0.26
107
+ },
108
+ {
109
+ "task": "arxiv_mc_2411",
110
+ "metric": "acc_stderr",
111
+ "value": 0.04408440022768079
112
+ },
113
+ {
114
+ "task": "arxiv_mc_2411",
115
+ "metric": "acc_norm",
116
+ "value": 0.4
117
+ },
118
+ {
119
+ "task": "arxiv_mc_2411",
120
+ "metric": "acc_norm_stderr",
121
+ "value": 0.049236596391733084
122
+ },
123
+ {
124
+ "task": "arxiv_mc_2412",
125
+ "metric": "acc",
126
+ "value": 0.12
127
+ },
128
+ {
129
+ "task": "arxiv_mc_2412",
130
+ "metric": "acc_stderr",
131
+ "value": 0.03265986323710906
132
+ },
133
+ {
134
+ "task": "arxiv_mc_2412",
135
+ "metric": "acc_norm",
136
+ "value": 0.22
137
+ },
138
+ {
139
+ "task": "arxiv_mc_2412",
140
+ "metric": "acc_norm_stderr",
141
+ "value": 0.04163331998932268
142
+ },
143
+ {
144
+ "task": "arxiv_mc_2501",
145
+ "metric": "acc",
146
+ "value": 0.18
147
+ },
148
+ {
149
+ "task": "arxiv_mc_2501",
150
+ "metric": "acc_stderr",
151
+ "value": 0.03861229196653695
152
+ },
153
+ {
154
+ "task": "arxiv_mc_2501",
155
+ "metric": "acc_norm",
156
+ "value": 0.23
157
+ },
158
+ {
159
+ "task": "arxiv_mc_2501",
160
+ "metric": "acc_norm_stderr",
161
+ "value": 0.04229525846816506
162
+ },
163
+ {
164
+ "task": "arxiv_mc_2502",
165
+ "metric": "acc",
166
+ "value": 0.2
167
+ },
168
+ {
169
+ "task": "arxiv_mc_2502",
170
+ "metric": "acc_stderr",
171
+ "value": 0.04020151261036844
172
+ },
173
+ {
174
+ "task": "arxiv_mc_2502",
175
+ "metric": "acc_norm",
176
+ "value": 0.26
177
+ },
178
+ {
179
+ "task": "arxiv_mc_2502",
180
+ "metric": "acc_norm_stderr",
181
+ "value": 0.04408440022768078
182
+ },
183
+ {
184
+ "task": "arxiv_mc_2503",
185
+ "metric": "acc",
186
+ "value": 0.2
187
+ },
188
+ {
189
+ "task": "arxiv_mc_2503",
190
+ "metric": "acc_stderr",
191
+ "value": 0.04020151261036846
192
+ },
193
+ {
194
+ "task": "arxiv_mc_2503",
195
+ "metric": "acc_norm",
196
+ "value": 0.27
197
+ },
198
+ {
199
+ "task": "arxiv_mc_2503",
200
+ "metric": "acc_norm_stderr",
201
+ "value": 0.04461960433384741
202
+ },
203
+ {
204
+ "task": "arxiv_mc_2504",
205
+ "metric": "acc",
206
+ "value": 0.11
207
+ },
208
+ {
209
+ "task": "arxiv_mc_2504",
210
+ "metric": "acc_stderr",
211
+ "value": 0.03144660377352203
212
+ },
213
+ {
214
+ "task": "arxiv_mc_2504",
215
+ "metric": "acc_norm",
216
+ "value": 0.2
217
+ },
218
+ {
219
+ "task": "arxiv_mc_2504",
220
+ "metric": "acc_norm_stderr",
221
+ "value": 0.04020151261036843
222
+ },
223
+ {
224
+ "task": "arxiv_mc_2505",
225
+ "metric": "acc",
226
+ "value": 0.18
227
+ },
228
+ {
229
+ "task": "arxiv_mc_2505",
230
+ "metric": "acc_stderr",
231
+ "value": 0.038612291966536955
232
+ },
233
+ {
234
+ "task": "arxiv_mc_2505",
235
+ "metric": "acc_norm",
236
+ "value": 0.21
237
+ },
238
+ {
239
+ "task": "arxiv_mc_2505",
240
+ "metric": "acc_norm_stderr",
241
+ "value": 0.040936018074033256
242
+ },
243
+ {
244
+ "task": "arxiv_mc_2506",
245
+ "metric": "acc",
246
+ "value": 0.18
247
+ },
248
+ {
249
+ "task": "arxiv_mc_2506",
250
+ "metric": "acc_stderr",
251
+ "value": 0.03861229196653695
252
+ },
253
+ {
254
+ "task": "arxiv_mc_2506",
255
+ "metric": "acc_norm",
256
+ "value": 0.36
257
+ },
258
+ {
259
+ "task": "arxiv_mc_2506",
260
+ "metric": "acc_norm_stderr",
261
+ "value": 0.048241815132442176
262
+ }
263
+ ],
264
+ "groups": [
265
+ {
266
+ "group": "arxiv_mc",
267
+ "metric": "acc",
268
+ "value": 0.17666666666666667
269
+ },
270
+ {
271
+ "group": "arxiv_mc",
272
+ "metric": "acc_stderr",
273
+ "value": 0.011004539416375157
274
+ },
275
+ {
276
+ "group": "arxiv_mc",
277
+ "metric": "acc_norm",
278
+ "value": 0.265
279
+ },
280
+ {
281
+ "group": "arxiv_mc",
282
+ "metric": "acc_norm_stderr",
283
+ "value": 0.012674705051722038
284
+ }
285
+ ]
286
+ }
out/eval/openllama_arxiv_mc/2411/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_arxiv_mc/2411/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_arxiv_mc/2411/log.txt ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/100 [00:00<?, ?it/s]
 
1
  0%| | 0/100 [00:00<?, ?it/s]
 
2
  0%| | 0/100 [00:00<?, ?it/s]
 
3
  0%| | 0/100 [00:00<?, ?it/s]
 
4
  0%| | 0/100 [00:00<?, ?it/s]
 
5
  0%| | 0/100 [00:00<?, ?it/s]
 
6
  0%| | 0/100 [00:00<?, ?it/s]
 
7
  0%| | 0/100 [00:00<?, ?it/s]
 
8
  0%| | 0/100 [00:00<?, ?it/s]
 
9
  0%| | 0/100 [00:00<?, ?it/s]
 
10
  0%| | 0/100 [00:00<?, ?it/s]
 
11
  0%| | 0/100 [00:00<?, ?it/s]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'access_token': None,
2
+ 'batch_size': 4,
3
+ 'checkpoint_dir': PosixPath('out/pretrain/openllama/2411/final'),
4
+ 'device': None,
5
+ 'dtype': None,
6
+ 'force_conversion': False,
7
+ 'limit': None,
8
+ 'num_fewshot': None,
9
+ 'out_dir': PosixPath('out/eval/openllama_arxiv_mc/2411'),
10
+ 'save_filepath': None,
11
+ 'seed': 1234,
12
+ 'tasks': 'arxiv_mc'}
13
+ INFO 09-24 15:04:51 [__init__.py:241] Automatically detected platform cuda.
14
+ Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
15
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message
16
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama_fast.LlamaTokenizerFast'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message.
17
+ {'checkpoint_dir': PosixPath('out/pretrain/openllama/2411/final'),
18
+ 'output_dir': PosixPath('out/eval/openllama_arxiv_mc/2411')}
19
+
20
  0%| | 0/100 [00:00<?, ?it/s]
21
+
22
  0%| | 0/100 [00:00<?, ?it/s]
23
+
24
  0%| | 0/100 [00:00<?, ?it/s]
25
+
26
  0%| | 0/100 [00:00<?, ?it/s]
27
+
28
  0%| | 0/100 [00:00<?, ?it/s]
29
+
30
  0%| | 0/100 [00:00<?, ?it/s]
31
+
32
  0%| | 0/100 [00:00<?, ?it/s]
33
+
34
  0%| | 0/100 [00:00<?, ?it/s]
35
+
36
  0%| | 0/100 [00:00<?, ?it/s]
37
+
38
  0%| | 0/100 [00:00<?, ?it/s]
39
+
40
  0%| | 0/100 [00:00<?, ?it/s]
41
+
42
  0%| | 0/100 [00:00<?, ?it/s]
43
+
44
+ | Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
45
+ |----------|-------|------|-----:|--------|---|-----:|---|-----:|
46
+ |arxiv_mc | |none | |acc |↑ |0.1967|± |0.0115|
47
+ | | |none | |acc_norm|↑ |0.2808|± |0.0129|
48
+ | - 2024-07|Yaml |none | 0|acc |↑ |0.2200|± |0.0416|
49
+ | | |none | 0|acc_norm|↑ |0.3000|± |0.0461|
50
+ | - 2024-08|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
51
+ | | |none | 0|acc_norm|↑ |0.2200|± |0.0416|
52
+ | - 2024-09|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
53
+ | | |none | 0|acc_norm|↑ |0.3300|± |0.0473|
54
+ | - 2024-10|Yaml |none | 0|acc |↑ |0.1300|± |0.0338|
55
+ | | |none | 0|acc_norm|↑ |0.2200|± |0.0416|
56
+ | - 2024-11|Yaml |none | 0|acc |↑ |0.2900|± |0.0456|
57
+ | | |none | 0|acc_norm|↑ |0.4100|± |0.0494|
58
+ | - 2024-12|Yaml |none | 0|acc |↑ |0.1500|± |0.0359|
59
+ | | |none | 0|acc_norm|↑ |0.2900|± |0.0456|
60
+ | - 2025-01|Yaml |none | 0|acc |↑ |0.1900|± |0.0394|
61
+ | | |none | 0|acc_norm|↑ |0.2500|± |0.0435|
62
+ | - 2025-02|Yaml |none | 0|acc |↑ |0.2200|± |0.0416|
63
+ | | |none | 0|acc_norm|↑ |0.3200|± |0.0469|
64
+ | - 2025-03|Yaml |none | 0|acc |↑ |0.2200|± |0.0416|
65
+ | | |none | 0|acc_norm|↑ |0.3000|± |0.0461|
66
+ | - 2025-04|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
67
+ | | |none | 0|acc_norm|↑ |0.1800|± |0.0386|
68
+ | - 2025-05|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
69
+ | | |none | 0|acc_norm|↑ |0.2400|± |0.0429|
70
+ | - 2025-06|Yaml |none | 0|acc |↑ |0.1900|± |0.0394|
71
+ | | |none | 0|acc_norm|↑ |0.3100|± |0.0465|
72
+
73
+ | Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
74
+ |--------|-------|------|------|--------|---|-----:|---|-----:|
75
+ |arxiv_mc| |none | |acc |↑ |0.1967|± |0.0115|
76
+ | | |none | |acc_norm|↑ |0.2808|± |0.0129|
77
+
out/eval/openllama_arxiv_mc/2411/model_config.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ attention_logit_softcapping: null
2
+ attention_scores_scalar: null
3
+ attn_bias: false
4
+ bias: false
5
+ block_size: 2048
6
+ final_logit_softcapping: null
7
+ gelu_approximate: none
8
+ head_size: 100
9
+ hf_config:
10
+ name: open_llama_3b
11
+ org: openlm-research
12
+ intermediate_size: 8640
13
+ lm_head_bias: false
14
+ mlp_class_name: LLaMAMLP
15
+ moe_intermediate_size: null
16
+ n_embd: 3200
17
+ n_expert: 0
18
+ n_expert_per_token: 0
19
+ n_head: 32
20
+ n_layer: 26
21
+ n_query_groups: 32
22
+ name: open_llama_3b
23
+ norm_1: true
24
+ norm_2: true
25
+ norm_class_name: RMSNorm
26
+ norm_eps: 1.0e-06
27
+ norm_qk: false
28
+ norm_qk_type: default
29
+ padded_vocab_size: 32000
30
+ padding_multiple: 64
31
+ parallel_residual: false
32
+ post_attention_norm: false
33
+ post_mlp_norm: false
34
+ rope_adjustments: null
35
+ rope_base: 10000
36
+ rope_condense_ratio: 1
37
+ rope_indices: null
38
+ rope_local_base_freq: null
39
+ rotary_percentage: 1.0
40
+ scale_embeddings: false
41
+ shared_attention_norm: false
42
+ sliding_window_indices: null
43
+ sliding_window_size: null
44
+ vocab_size: 32000
out/eval/openllama_arxiv_mc/2411/results.json ADDED
The diff for this file is too large to render. See raw diff
 
out/eval/openllama_arxiv_mc/2411/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"add_bos_token": true, "add_eos_token": false, "model_max_length": 2048, "pad_token": null, "sp_model_kwargs": {}, "tokenizer_class": "LlamaTokenizer", "clean_up_tokenization_spaces": false, "bos_token": {"__type": "AddedToken", "content": "<s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "eos_token": {"__type": "AddedToken", "content": "</s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "unk_token": {"__type": "AddedToken", "content": "<unk>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}}
out/eval/openllama_arxiv_mc/2411/values.json ADDED
@@ -0,0 +1,286 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tasks": [
3
+ {
4
+ "task": "arxiv_mc",
5
+ "metric": "acc",
6
+ "value": 0.19666666666666666
7
+ },
8
+ {
9
+ "task": "arxiv_mc",
10
+ "metric": "acc_stderr",
11
+ "value": 0.01147693241041834
12
+ },
13
+ {
14
+ "task": "arxiv_mc",
15
+ "metric": "acc_norm",
16
+ "value": 0.2808333333333333
17
+ },
18
+ {
19
+ "task": "arxiv_mc",
20
+ "metric": "acc_norm_stderr",
21
+ "value": 0.01292379243002652
22
+ },
23
+ {
24
+ "task": "arxiv_mc_2407",
25
+ "metric": "acc",
26
+ "value": 0.22
27
+ },
28
+ {
29
+ "task": "arxiv_mc_2407",
30
+ "metric": "acc_stderr",
31
+ "value": 0.04163331998932269
32
+ },
33
+ {
34
+ "task": "arxiv_mc_2407",
35
+ "metric": "acc_norm",
36
+ "value": 0.3
37
+ },
38
+ {
39
+ "task": "arxiv_mc_2407",
40
+ "metric": "acc_norm_stderr",
41
+ "value": 0.046056618647183814
42
+ },
43
+ {
44
+ "task": "arxiv_mc_2408",
45
+ "metric": "acc",
46
+ "value": 0.18
47
+ },
48
+ {
49
+ "task": "arxiv_mc_2408",
50
+ "metric": "acc_stderr",
51
+ "value": 0.038612291966536955
52
+ },
53
+ {
54
+ "task": "arxiv_mc_2408",
55
+ "metric": "acc_norm",
56
+ "value": 0.22
57
+ },
58
+ {
59
+ "task": "arxiv_mc_2408",
60
+ "metric": "acc_norm_stderr",
61
+ "value": 0.0416333199893227
62
+ },
63
+ {
64
+ "task": "arxiv_mc_2409",
65
+ "metric": "acc",
66
+ "value": 0.2
67
+ },
68
+ {
69
+ "task": "arxiv_mc_2409",
70
+ "metric": "acc_stderr",
71
+ "value": 0.040201512610368445
72
+ },
73
+ {
74
+ "task": "arxiv_mc_2409",
75
+ "metric": "acc_norm",
76
+ "value": 0.33
77
+ },
78
+ {
79
+ "task": "arxiv_mc_2409",
80
+ "metric": "acc_norm_stderr",
81
+ "value": 0.047258156262526045
82
+ },
83
+ {
84
+ "task": "arxiv_mc_2410",
85
+ "metric": "acc",
86
+ "value": 0.13
87
+ },
88
+ {
89
+ "task": "arxiv_mc_2410",
90
+ "metric": "acc_stderr",
91
+ "value": 0.033799766898963086
92
+ },
93
+ {
94
+ "task": "arxiv_mc_2410",
95
+ "metric": "acc_norm",
96
+ "value": 0.22
97
+ },
98
+ {
99
+ "task": "arxiv_mc_2410",
100
+ "metric": "acc_norm_stderr",
101
+ "value": 0.041633319989322695
102
+ },
103
+ {
104
+ "task": "arxiv_mc_2411",
105
+ "metric": "acc",
106
+ "value": 0.29
107
+ },
108
+ {
109
+ "task": "arxiv_mc_2411",
110
+ "metric": "acc_stderr",
111
+ "value": 0.045604802157206845
112
+ },
113
+ {
114
+ "task": "arxiv_mc_2411",
115
+ "metric": "acc_norm",
116
+ "value": 0.41
117
+ },
118
+ {
119
+ "task": "arxiv_mc_2411",
120
+ "metric": "acc_norm_stderr",
121
+ "value": 0.049431107042371025
122
+ },
123
+ {
124
+ "task": "arxiv_mc_2412",
125
+ "metric": "acc",
126
+ "value": 0.15
127
+ },
128
+ {
129
+ "task": "arxiv_mc_2412",
130
+ "metric": "acc_stderr",
131
+ "value": 0.035887028128263714
132
+ },
133
+ {
134
+ "task": "arxiv_mc_2412",
135
+ "metric": "acc_norm",
136
+ "value": 0.29
137
+ },
138
+ {
139
+ "task": "arxiv_mc_2412",
140
+ "metric": "acc_norm_stderr",
141
+ "value": 0.045604802157206845
142
+ },
143
+ {
144
+ "task": "arxiv_mc_2501",
145
+ "metric": "acc",
146
+ "value": 0.19
147
+ },
148
+ {
149
+ "task": "arxiv_mc_2501",
150
+ "metric": "acc_stderr",
151
+ "value": 0.03942772444036623
152
+ },
153
+ {
154
+ "task": "arxiv_mc_2501",
155
+ "metric": "acc_norm",
156
+ "value": 0.25
157
+ },
158
+ {
159
+ "task": "arxiv_mc_2501",
160
+ "metric": "acc_norm_stderr",
161
+ "value": 0.04351941398892446
162
+ },
163
+ {
164
+ "task": "arxiv_mc_2502",
165
+ "metric": "acc",
166
+ "value": 0.22
167
+ },
168
+ {
169
+ "task": "arxiv_mc_2502",
170
+ "metric": "acc_stderr",
171
+ "value": 0.041633319989322695
172
+ },
173
+ {
174
+ "task": "arxiv_mc_2502",
175
+ "metric": "acc_norm",
176
+ "value": 0.32
177
+ },
178
+ {
179
+ "task": "arxiv_mc_2502",
180
+ "metric": "acc_norm_stderr",
181
+ "value": 0.04688261722621504
182
+ },
183
+ {
184
+ "task": "arxiv_mc_2503",
185
+ "metric": "acc",
186
+ "value": 0.22
187
+ },
188
+ {
189
+ "task": "arxiv_mc_2503",
190
+ "metric": "acc_stderr",
191
+ "value": 0.04163331998932269
192
+ },
193
+ {
194
+ "task": "arxiv_mc_2503",
195
+ "metric": "acc_norm",
196
+ "value": 0.3
197
+ },
198
+ {
199
+ "task": "arxiv_mc_2503",
200
+ "metric": "acc_norm_stderr",
201
+ "value": 0.046056618647183814
202
+ },
203
+ {
204
+ "task": "arxiv_mc_2504",
205
+ "metric": "acc",
206
+ "value": 0.17
207
+ },
208
+ {
209
+ "task": "arxiv_mc_2504",
210
+ "metric": "acc_stderr",
211
+ "value": 0.0377525168068637
212
+ },
213
+ {
214
+ "task": "arxiv_mc_2504",
215
+ "metric": "acc_norm",
216
+ "value": 0.18
217
+ },
218
+ {
219
+ "task": "arxiv_mc_2504",
220
+ "metric": "acc_norm_stderr",
221
+ "value": 0.03861229196653697
222
+ },
223
+ {
224
+ "task": "arxiv_mc_2505",
225
+ "metric": "acc",
226
+ "value": 0.2
227
+ },
228
+ {
229
+ "task": "arxiv_mc_2505",
230
+ "metric": "acc_stderr",
231
+ "value": 0.040201512610368445
232
+ },
233
+ {
234
+ "task": "arxiv_mc_2505",
235
+ "metric": "acc_norm",
236
+ "value": 0.24
237
+ },
238
+ {
239
+ "task": "arxiv_mc_2505",
240
+ "metric": "acc_norm_stderr",
241
+ "value": 0.042923469599092816
242
+ },
243
+ {
244
+ "task": "arxiv_mc_2506",
245
+ "metric": "acc",
246
+ "value": 0.19
247
+ },
248
+ {
249
+ "task": "arxiv_mc_2506",
250
+ "metric": "acc_stderr",
251
+ "value": 0.039427724440366234
252
+ },
253
+ {
254
+ "task": "arxiv_mc_2506",
255
+ "metric": "acc_norm",
256
+ "value": 0.31
257
+ },
258
+ {
259
+ "task": "arxiv_mc_2506",
260
+ "metric": "acc_norm_stderr",
261
+ "value": 0.04648231987117316
262
+ }
263
+ ],
264
+ "groups": [
265
+ {
266
+ "group": "arxiv_mc",
267
+ "metric": "acc",
268
+ "value": 0.19666666666666666
269
+ },
270
+ {
271
+ "group": "arxiv_mc",
272
+ "metric": "acc_stderr",
273
+ "value": 0.01147693241041834
274
+ },
275
+ {
276
+ "group": "arxiv_mc",
277
+ "metric": "acc_norm",
278
+ "value": 0.2808333333333333
279
+ },
280
+ {
281
+ "group": "arxiv_mc",
282
+ "metric": "acc_norm_stderr",
283
+ "value": 0.01292379243002652
284
+ }
285
+ ]
286
+ }
out/eval/openllama_arxiv_mc/2412/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_arxiv_mc/2412/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_arxiv_mc/2412/log.txt ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/100 [00:00<?, ?it/s]
 
1
  0%| | 0/100 [00:00<?, ?it/s]
 
2
  0%| | 0/100 [00:00<?, ?it/s]
 
3
  0%| | 0/100 [00:00<?, ?it/s]
 
4
  0%| | 0/100 [00:00<?, ?it/s]
 
5
  0%| | 0/100 [00:00<?, ?it/s]
 
6
  0%| | 0/100 [00:00<?, ?it/s]
 
7
  0%| | 0/100 [00:00<?, ?it/s]
 
8
  0%| | 0/100 [00:00<?, ?it/s]
 
9
  0%| | 0/100 [00:00<?, ?it/s]
 
10
  0%| | 0/100 [00:00<?, ?it/s]
 
11
  0%| | 0/100 [00:00<?, ?it/s]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'access_token': None,
2
+ 'batch_size': 4,
3
+ 'checkpoint_dir': PosixPath('out/pretrain/openllama/2412/final'),
4
+ 'device': None,
5
+ 'dtype': None,
6
+ 'force_conversion': False,
7
+ 'limit': None,
8
+ 'num_fewshot': None,
9
+ 'out_dir': PosixPath('out/eval/openllama_arxiv_mc/2412'),
10
+ 'save_filepath': None,
11
+ 'seed': 1234,
12
+ 'tasks': 'arxiv_mc'}
13
+ INFO 09-24 15:06:39 [__init__.py:241] Automatically detected platform cuda.
14
+ Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
15
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message
16
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama_fast.LlamaTokenizerFast'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message.
17
+ {'checkpoint_dir': PosixPath('out/pretrain/openllama/2412/final'),
18
+ 'output_dir': PosixPath('out/eval/openllama_arxiv_mc/2412')}
19
+
20
  0%| | 0/100 [00:00<?, ?it/s]
21
+
22
  0%| | 0/100 [00:00<?, ?it/s]
23
+
24
  0%| | 0/100 [00:00<?, ?it/s]
25
+
26
  0%| | 0/100 [00:00<?, ?it/s]
27
+
28
  0%| | 0/100 [00:00<?, ?it/s]
29
+
30
  0%| | 0/100 [00:00<?, ?it/s]
31
+
32
  0%| | 0/100 [00:00<?, ?it/s]
33
+
34
  0%| | 0/100 [00:00<?, ?it/s]
35
+
36
  0%| | 0/100 [00:00<?, ?it/s]
37
+
38
  0%| | 0/100 [00:00<?, ?it/s]
39
+
40
  0%| | 0/100 [00:00<?, ?it/s]
41
+
42
  0%| | 0/100 [00:00<?, ?it/s]
43
+
44
+ | Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
45
+ |----------|-------|------|-----:|--------|---|-----:|---|-----:|
46
+ |arxiv_mc | |none | |acc |↑ |0.2025|± |0.0116|
47
+ | | |none | |acc_norm|↑ |0.2808|± |0.0129|
48
+ | - 2024-07|Yaml |none | 0|acc |↑ |0.2300|± |0.0423|
49
+ | | |none | 0|acc_norm|↑ |0.3400|± |0.0476|
50
+ | - 2024-08|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
51
+ | | |none | 0|acc_norm|↑ |0.2000|± |0.0402|
52
+ | - 2024-09|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
53
+ | | |none | 0|acc_norm|↑ |0.2900|± |0.0456|
54
+ | - 2024-10|Yaml |none | 0|acc |↑ |0.1400|± |0.0349|
55
+ | | |none | 0|acc_norm|↑ |0.2600|± |0.0441|
56
+ | - 2024-11|Yaml |none | 0|acc |↑ |0.2800|± |0.0451|
57
+ | | |none | 0|acc_norm|↑ |0.4100|± |0.0494|
58
+ | - 2024-12|Yaml |none | 0|acc |↑ |0.1900|± |0.0394|
59
+ | | |none | 0|acc_norm|↑ |0.2900|± |0.0456|
60
+ | - 2025-01|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
61
+ | | |none | 0|acc_norm|↑ |0.2400|± |0.0429|
62
+ | - 2025-02|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
63
+ | | |none | 0|acc_norm|↑ |0.3000|± |0.0461|
64
+ | - 2025-03|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
65
+ | | |none | 0|acc_norm|↑ |0.2700|± |0.0446|
66
+ | - 2025-04|Yaml |none | 0|acc |↑ |0.1500|± |0.0359|
67
+ | | |none | 0|acc_norm|↑ |0.2000|± |0.0402|
68
+ | - 2025-05|Yaml |none | 0|acc |↑ |0.2400|± |0.0429|
69
+ | | |none | 0|acc_norm|↑ |0.2500|± |0.0435|
70
+ | - 2025-06|Yaml |none | 0|acc |↑ |0.1800|± |0.0386|
71
+ | | |none | 0|acc_norm|↑ |0.3200|± |0.0469|
72
+
73
+ | Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
74
+ |--------|-------|------|------|--------|---|-----:|---|-----:|
75
+ |arxiv_mc| |none | |acc |↑ |0.2025|± |0.0116|
76
+ | | |none | |acc_norm|↑ |0.2808|± |0.0129|
77
+
out/eval/openllama_arxiv_mc/2412/model_config.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ attention_logit_softcapping: null
2
+ attention_scores_scalar: null
3
+ attn_bias: false
4
+ bias: false
5
+ block_size: 2048
6
+ final_logit_softcapping: null
7
+ gelu_approximate: none
8
+ head_size: 100
9
+ hf_config:
10
+ name: open_llama_3b
11
+ org: openlm-research
12
+ intermediate_size: 8640
13
+ lm_head_bias: false
14
+ mlp_class_name: LLaMAMLP
15
+ moe_intermediate_size: null
16
+ n_embd: 3200
17
+ n_expert: 0
18
+ n_expert_per_token: 0
19
+ n_head: 32
20
+ n_layer: 26
21
+ n_query_groups: 32
22
+ name: open_llama_3b
23
+ norm_1: true
24
+ norm_2: true
25
+ norm_class_name: RMSNorm
26
+ norm_eps: 1.0e-06
27
+ norm_qk: false
28
+ norm_qk_type: default
29
+ padded_vocab_size: 32000
30
+ padding_multiple: 64
31
+ parallel_residual: false
32
+ post_attention_norm: false
33
+ post_mlp_norm: false
34
+ rope_adjustments: null
35
+ rope_base: 10000
36
+ rope_condense_ratio: 1
37
+ rope_indices: null
38
+ rope_local_base_freq: null
39
+ rotary_percentage: 1.0
40
+ scale_embeddings: false
41
+ shared_attention_norm: false
42
+ sliding_window_indices: null
43
+ sliding_window_size: null
44
+ vocab_size: 32000
out/eval/openllama_arxiv_mc/2412/results.json ADDED
The diff for this file is too large to render. See raw diff
 
out/eval/openllama_arxiv_mc/2412/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"add_bos_token": true, "add_eos_token": false, "model_max_length": 2048, "pad_token": null, "sp_model_kwargs": {}, "tokenizer_class": "LlamaTokenizer", "clean_up_tokenization_spaces": false, "bos_token": {"__type": "AddedToken", "content": "<s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "eos_token": {"__type": "AddedToken", "content": "</s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "unk_token": {"__type": "AddedToken", "content": "<unk>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}}
out/eval/openllama_arxiv_mc/2412/values.json ADDED
@@ -0,0 +1,286 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tasks": [
3
+ {
4
+ "task": "arxiv_mc",
5
+ "metric": "acc_norm",
6
+ "value": 0.2808333333333333
7
+ },
8
+ {
9
+ "task": "arxiv_mc",
10
+ "metric": "acc_norm_stderr",
11
+ "value": 0.012934643187514978
12
+ },
13
+ {
14
+ "task": "arxiv_mc",
15
+ "metric": "acc",
16
+ "value": 0.2025
17
+ },
18
+ {
19
+ "task": "arxiv_mc",
20
+ "metric": "acc_stderr",
21
+ "value": 0.011612126000786286
22
+ },
23
+ {
24
+ "task": "arxiv_mc_2407",
25
+ "metric": "acc",
26
+ "value": 0.23
27
+ },
28
+ {
29
+ "task": "arxiv_mc_2407",
30
+ "metric": "acc_stderr",
31
+ "value": 0.04229525846816506
32
+ },
33
+ {
34
+ "task": "arxiv_mc_2407",
35
+ "metric": "acc_norm",
36
+ "value": 0.34
37
+ },
38
+ {
39
+ "task": "arxiv_mc_2407",
40
+ "metric": "acc_norm_stderr",
41
+ "value": 0.04760952285695235
42
+ },
43
+ {
44
+ "task": "arxiv_mc_2408",
45
+ "metric": "acc",
46
+ "value": 0.2
47
+ },
48
+ {
49
+ "task": "arxiv_mc_2408",
50
+ "metric": "acc_stderr",
51
+ "value": 0.040201512610368445
52
+ },
53
+ {
54
+ "task": "arxiv_mc_2408",
55
+ "metric": "acc_norm",
56
+ "value": 0.2
57
+ },
58
+ {
59
+ "task": "arxiv_mc_2408",
60
+ "metric": "acc_norm_stderr",
61
+ "value": 0.04020151261036844
62
+ },
63
+ {
64
+ "task": "arxiv_mc_2409",
65
+ "metric": "acc",
66
+ "value": 0.21
67
+ },
68
+ {
69
+ "task": "arxiv_mc_2409",
70
+ "metric": "acc_stderr",
71
+ "value": 0.040936018074033256
72
+ },
73
+ {
74
+ "task": "arxiv_mc_2409",
75
+ "metric": "acc_norm",
76
+ "value": 0.29
77
+ },
78
+ {
79
+ "task": "arxiv_mc_2409",
80
+ "metric": "acc_norm_stderr",
81
+ "value": 0.045604802157206845
82
+ },
83
+ {
84
+ "task": "arxiv_mc_2410",
85
+ "metric": "acc",
86
+ "value": 0.14
87
+ },
88
+ {
89
+ "task": "arxiv_mc_2410",
90
+ "metric": "acc_stderr",
91
+ "value": 0.034873508801977704
92
+ },
93
+ {
94
+ "task": "arxiv_mc_2410",
95
+ "metric": "acc_norm",
96
+ "value": 0.26
97
+ },
98
+ {
99
+ "task": "arxiv_mc_2410",
100
+ "metric": "acc_norm_stderr",
101
+ "value": 0.04408440022768081
102
+ },
103
+ {
104
+ "task": "arxiv_mc_2411",
105
+ "metric": "acc",
106
+ "value": 0.28
107
+ },
108
+ {
109
+ "task": "arxiv_mc_2411",
110
+ "metric": "acc_stderr",
111
+ "value": 0.04512608598542127
112
+ },
113
+ {
114
+ "task": "arxiv_mc_2411",
115
+ "metric": "acc_norm",
116
+ "value": 0.41
117
+ },
118
+ {
119
+ "task": "arxiv_mc_2411",
120
+ "metric": "acc_norm_stderr",
121
+ "value": 0.049431107042371025
122
+ },
123
+ {
124
+ "task": "arxiv_mc_2412",
125
+ "metric": "acc",
126
+ "value": 0.19
127
+ },
128
+ {
129
+ "task": "arxiv_mc_2412",
130
+ "metric": "acc_stderr",
131
+ "value": 0.039427724440366234
132
+ },
133
+ {
134
+ "task": "arxiv_mc_2412",
135
+ "metric": "acc_norm",
136
+ "value": 0.29
137
+ },
138
+ {
139
+ "task": "arxiv_mc_2412",
140
+ "metric": "acc_norm_stderr",
141
+ "value": 0.045604802157206845
142
+ },
143
+ {
144
+ "task": "arxiv_mc_2501",
145
+ "metric": "acc",
146
+ "value": 0.2
147
+ },
148
+ {
149
+ "task": "arxiv_mc_2501",
150
+ "metric": "acc_stderr",
151
+ "value": 0.04020151261036843
152
+ },
153
+ {
154
+ "task": "arxiv_mc_2501",
155
+ "metric": "acc_norm",
156
+ "value": 0.24
157
+ },
158
+ {
159
+ "task": "arxiv_mc_2501",
160
+ "metric": "acc_norm_stderr",
161
+ "value": 0.04292346959909284
162
+ },
163
+ {
164
+ "task": "arxiv_mc_2502",
165
+ "metric": "acc",
166
+ "value": 0.2
167
+ },
168
+ {
169
+ "task": "arxiv_mc_2502",
170
+ "metric": "acc_stderr",
171
+ "value": 0.040201512610368445
172
+ },
173
+ {
174
+ "task": "arxiv_mc_2502",
175
+ "metric": "acc_norm",
176
+ "value": 0.3
177
+ },
178
+ {
179
+ "task": "arxiv_mc_2502",
180
+ "metric": "acc_norm_stderr",
181
+ "value": 0.046056618647183814
182
+ },
183
+ {
184
+ "task": "arxiv_mc_2503",
185
+ "metric": "acc",
186
+ "value": 0.21
187
+ },
188
+ {
189
+ "task": "arxiv_mc_2503",
190
+ "metric": "acc_stderr",
191
+ "value": 0.040936018074033256
192
+ },
193
+ {
194
+ "task": "arxiv_mc_2503",
195
+ "metric": "acc_norm",
196
+ "value": 0.27
197
+ },
198
+ {
199
+ "task": "arxiv_mc_2503",
200
+ "metric": "acc_norm_stderr",
201
+ "value": 0.04461960433384741
202
+ },
203
+ {
204
+ "task": "arxiv_mc_2504",
205
+ "metric": "acc",
206
+ "value": 0.15
207
+ },
208
+ {
209
+ "task": "arxiv_mc_2504",
210
+ "metric": "acc_stderr",
211
+ "value": 0.03588702812826371
212
+ },
213
+ {
214
+ "task": "arxiv_mc_2504",
215
+ "metric": "acc_norm",
216
+ "value": 0.2
217
+ },
218
+ {
219
+ "task": "arxiv_mc_2504",
220
+ "metric": "acc_norm_stderr",
221
+ "value": 0.040201512610368445
222
+ },
223
+ {
224
+ "task": "arxiv_mc_2505",
225
+ "metric": "acc",
226
+ "value": 0.24
227
+ },
228
+ {
229
+ "task": "arxiv_mc_2505",
230
+ "metric": "acc_stderr",
231
+ "value": 0.04292346959909283
232
+ },
233
+ {
234
+ "task": "arxiv_mc_2505",
235
+ "metric": "acc_norm",
236
+ "value": 0.25
237
+ },
238
+ {
239
+ "task": "arxiv_mc_2505",
240
+ "metric": "acc_norm_stderr",
241
+ "value": 0.04351941398892446
242
+ },
243
+ {
244
+ "task": "arxiv_mc_2506",
245
+ "metric": "acc",
246
+ "value": 0.18
247
+ },
248
+ {
249
+ "task": "arxiv_mc_2506",
250
+ "metric": "acc_stderr",
251
+ "value": 0.03861229196653694
252
+ },
253
+ {
254
+ "task": "arxiv_mc_2506",
255
+ "metric": "acc_norm",
256
+ "value": 0.32
257
+ },
258
+ {
259
+ "task": "arxiv_mc_2506",
260
+ "metric": "acc_norm_stderr",
261
+ "value": 0.04688261722621504
262
+ }
263
+ ],
264
+ "groups": [
265
+ {
266
+ "group": "arxiv_mc",
267
+ "metric": "acc_norm",
268
+ "value": 0.2808333333333333
269
+ },
270
+ {
271
+ "group": "arxiv_mc",
272
+ "metric": "acc_norm_stderr",
273
+ "value": 0.012934643187514978
274
+ },
275
+ {
276
+ "group": "arxiv_mc",
277
+ "metric": "acc",
278
+ "value": 0.2025
279
+ },
280
+ {
281
+ "group": "arxiv_mc",
282
+ "metric": "acc_stderr",
283
+ "value": 0.011612126000786286
284
+ }
285
+ ]
286
+ }
out/eval/openllama_arxiv_mc/2501/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_arxiv_mc/2501/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_arxiv_mc/2501/log.txt ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/100 [00:00<?, ?it/s]
 
1
  0%| | 0/100 [00:00<?, ?it/s]
 
2
  0%| | 0/100 [00:00<?, ?it/s]
 
3
  0%| | 0/100 [00:00<?, ?it/s]
 
4
  0%| | 0/100 [00:00<?, ?it/s]
 
5
  0%| | 0/100 [00:00<?, ?it/s]
 
6
  0%| | 0/100 [00:00<?, ?it/s]
 
7
  0%| | 0/100 [00:00<?, ?it/s]
 
8
  0%| | 0/100 [00:00<?, ?it/s]
 
9
  0%| | 0/100 [00:00<?, ?it/s]
 
10
  0%| | 0/100 [00:00<?, ?it/s]
 
11
  0%| | 0/100 [00:00<?, ?it/s]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'access_token': None,
2
+ 'batch_size': 4,
3
+ 'checkpoint_dir': PosixPath('out/pretrain/openllama/2501/final'),
4
+ 'device': None,
5
+ 'dtype': None,
6
+ 'force_conversion': False,
7
+ 'limit': None,
8
+ 'num_fewshot': None,
9
+ 'out_dir': PosixPath('out/eval/openllama_arxiv_mc/2501'),
10
+ 'save_filepath': None,
11
+ 'seed': 1234,
12
+ 'tasks': 'arxiv_mc'}
13
+ INFO 09-24 15:08:28 [__init__.py:241] Automatically detected platform cuda.
14
+ Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
15
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message
16
+ You are using the default legacy behaviour of the <class 'transformers.models.llama.tokenization_llama_fast.LlamaTokenizerFast'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 - if you loaded a llama tokenizer from a GGUF file you can ignore this message.
17
+ {'checkpoint_dir': PosixPath('out/pretrain/openllama/2501/final'),
18
+ 'output_dir': PosixPath('out/eval/openllama_arxiv_mc/2501')}
19
+
20
  0%| | 0/100 [00:00<?, ?it/s]
21
+
22
  0%| | 0/100 [00:00<?, ?it/s]
23
+
24
  0%| | 0/100 [00:00<?, ?it/s]
25
+
26
  0%| | 0/100 [00:00<?, ?it/s]
27
+
28
  0%| | 0/100 [00:00<?, ?it/s]
29
+
30
  0%| | 0/100 [00:00<?, ?it/s]
31
+
32
  0%| | 0/100 [00:00<?, ?it/s]
33
+
34
  0%| | 0/100 [00:00<?, ?it/s]
35
+
36
  0%| | 0/100 [00:00<?, ?it/s]
37
+
38
  0%| | 0/100 [00:00<?, ?it/s]
39
+
40
  0%| | 0/100 [00:00<?, ?it/s]
41
+
42
  0%| | 0/100 [00:00<?, ?it/s]
43
+
44
+ | Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
45
+ |----------|-------|------|-----:|--------|---|-----:|---|-----:|
46
+ |arxiv_mc | |none | |acc |↑ |0.1975|± |0.0115|
47
+ | | |none | |acc_norm|↑ |0.2908|± |0.0131|
48
+ | - 2024-07|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
49
+ | | |none | 0|acc_norm|↑ |0.3200|± |0.0469|
50
+ | - 2024-08|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
51
+ | | |none | 0|acc_norm|↑ |0.2600|± |0.0441|
52
+ | - 2024-09|Yaml |none | 0|acc |↑ |0.2000|± |0.0402|
53
+ | | |none | 0|acc_norm|↑ |0.2900|± |0.0456|
54
+ | - 2024-10|Yaml |none | 0|acc |↑ |0.1200|± |0.0327|
55
+ | | |none | 0|acc_norm|↑ |0.2600|± |0.0441|
56
+ | - 2024-11|Yaml |none | 0|acc |↑ |0.3000|± |0.0461|
57
+ | | |none | 0|acc_norm|↑ |0.4100|± |0.0494|
58
+ | - 2024-12|Yaml |none | 0|acc |↑ |0.1600|± |0.0368|
59
+ | | |none | 0|acc_norm|↑ |0.2700|± |0.0446|
60
+ | - 2025-01|Yaml |none | 0|acc |↑ |0.2600|± |0.0441|
61
+ | | |none | 0|acc_norm|↑ |0.2900|± |0.0456|
62
+ | - 2025-02|Yaml |none | 0|acc |↑ |0.2500|± |0.0435|
63
+ | | |none | 0|acc_norm|↑ |0.3200|± |0.0469|
64
+ | - 2025-03|Yaml |none | 0|acc |↑ |0.1700|± |0.0378|
65
+ | | |none | 0|acc_norm|↑ |0.2400|± |0.0429|
66
+ | - 2025-04|Yaml |none | 0|acc |↑ |0.1400|± |0.0349|
67
+ | | |none | 0|acc_norm|↑ |0.2500|± |0.0435|
68
+ | - 2025-05|Yaml |none | 0|acc |↑ |0.2100|± |0.0409|
69
+ | | |none | 0|acc_norm|↑ |0.2600|± |0.0441|
70
+ | - 2025-06|Yaml |none | 0|acc |↑ |0.1900|± |0.0394|
71
+ | | |none | 0|acc_norm|↑ |0.3200|± |0.0469|
72
+
73
+ | Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
74
+ |--------|-------|------|------|--------|---|-----:|---|-----:|
75
+ |arxiv_mc| |none | |acc |↑ |0.1975|± |0.0115|
76
+ | | |none | |acc_norm|↑ |0.2908|± |0.0131|
77
+
out/eval/openllama_arxiv_mc/2501/model_config.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ attention_logit_softcapping: null
2
+ attention_scores_scalar: null
3
+ attn_bias: false
4
+ bias: false
5
+ block_size: 2048
6
+ final_logit_softcapping: null
7
+ gelu_approximate: none
8
+ head_size: 100
9
+ hf_config:
10
+ name: open_llama_3b
11
+ org: openlm-research
12
+ intermediate_size: 8640
13
+ lm_head_bias: false
14
+ mlp_class_name: LLaMAMLP
15
+ moe_intermediate_size: null
16
+ n_embd: 3200
17
+ n_expert: 0
18
+ n_expert_per_token: 0
19
+ n_head: 32
20
+ n_layer: 26
21
+ n_query_groups: 32
22
+ name: open_llama_3b
23
+ norm_1: true
24
+ norm_2: true
25
+ norm_class_name: RMSNorm
26
+ norm_eps: 1.0e-06
27
+ norm_qk: false
28
+ norm_qk_type: default
29
+ padded_vocab_size: 32000
30
+ padding_multiple: 64
31
+ parallel_residual: false
32
+ post_attention_norm: false
33
+ post_mlp_norm: false
34
+ rope_adjustments: null
35
+ rope_base: 10000
36
+ rope_condense_ratio: 1
37
+ rope_indices: null
38
+ rope_local_base_freq: null
39
+ rotary_percentage: 1.0
40
+ scale_embeddings: false
41
+ shared_attention_norm: false
42
+ sliding_window_indices: null
43
+ sliding_window_size: null
44
+ vocab_size: 32000
out/eval/openllama_arxiv_mc/2501/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"add_bos_token": true, "add_eos_token": false, "model_max_length": 2048, "pad_token": null, "sp_model_kwargs": {}, "tokenizer_class": "LlamaTokenizer", "clean_up_tokenization_spaces": false, "bos_token": {"__type": "AddedToken", "content": "<s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "eos_token": {"__type": "AddedToken", "content": "</s>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}, "unk_token": {"__type": "AddedToken", "content": "<unk>", "lstrip": false, "normalized": true, "rstrip": false, "single_word": false}}
out/eval/openllama_benches/2407/config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 3200,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 8640,
11
+ "max_position_embeddings": 2048,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 32,
14
+ "num_hidden_layers": 26,
15
+ "pad_token_id": 0,
16
+ "rms_norm_eps": 1e-06,
17
+ "tie_word_embeddings": false,
18
+ "torch_dtype": "float16",
19
+ "transformers_version": "4.30.0.dev0",
20
+ "use_cache": true,
21
+ "vocab_size": 32000
22
+ }
out/eval/openllama_benches/2407/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.30.0.dev0"
7
+ }
out/eval/openllama_benches/2407/log.txt ADDED
The diff for this file is too large to render. See raw diff