runs/molab: logs
Browse files- runs/molab/logs/console.log +71 -225
- runs/molab/logs/pipeline.log +66 -201
runs/molab/logs/console.log
CHANGED
|
@@ -1,234 +1,80 @@
|
|
| 1 |
No files have been modified since last commit. Skipping to prevent empty commit.
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
[transformers] Could not extract SentencePiece model from /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9fdd2ebd8db0901d37b98d3a53bbfaa4503/tokenizer.model using sentencepiece library due to
|
| 56 |
-
SentencePieceExtractor requires the SentencePiece library but it was not found in your environment. Check out the instructions on the
|
| 57 |
-
installation page of its repo: https://github.com/google/sentencepiece#installation and follow the ones
|
| 58 |
-
that match your environment. Please note that you may need to restart your runtime after installation.
|
| 59 |
-
. Falling back to TikToken extractor.
|
| 60 |
-
15:04:53 | WARNING | could not load NousResearch/Llama-2-13b-hf: Error parsing line b'\x0e' in /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9fdd2ebd8db0901d37b98d3a53bbfaa4503/tokenizer.model
|
| 61 |
-
15:04:53 | ERROR | could not prepare model llama-13b: failed to load llama-13b: Error parsing line b'\x0e' in /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9fdd2ebd8db0901d37b98d3a53bbfaa4503/tokenizer.model
|
| 62 |
-
Traceback (most recent call last):
|
| 63 |
-
File "/home/marimo/hagdcodebase/src/hagd/pipeline.py", line 545, in _open
|
| 64 |
-
ctx = ModelContext(self, key, tasks)
|
| 65 |
-
File "/home/marimo/hagdcodebase/src/hagd/pipeline.py", line 58, in __init__
|
| 66 |
-
self.lm = LM.load(model_spec(pipe.cfg, key), pipe.device, self.cfg.get("quantization"))
|
| 67 |
-
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 68 |
-
File "/home/marimo/hagdcodebase/src/hagd/models.py", line 119, in load
|
| 69 |
-
raise RuntimeError(f"failed to load {spec.key}: {last_err}")
|
| 70 |
-
RuntimeError: failed to load llama-13b: Error parsing line b'\x0e' in /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9fdd2ebd8db0901d37b98d3a53bbfaa4503/tokenizer.model
|
| 71 |
-
|
| 72 |
-
15:04:53 | INFO | === phase: transfer ===
|
| 73 |
-
15:04:53 | INFO | hub: restored 2 file(s) for gpt2-large/circuits/mod_arith/hierarchical
|
| 74 |
-
15:04:55 | INFO | hub: restored 4 file(s) for gpt2-large/graphs/mod_arith
|
| 75 |
-
15:04:55 | INFO | hub: restored 2 file(s) for pythia-2.8b/circuits/mod_arith/hierarchical
|
| 76 |
-
15:04:56 | INFO | hub: restored 4 file(s) for pythia-2.8b/graphs/mod_arith
|
| 77 |
-
15:04:58 | INFO | transfer gpt2-large -> pythia-2.8b on mod_arith: tau = 0.057 (767 aligned features)
|
| 78 |
-
15:04:58 | INFO | hub: restored 2 file(s) for gpt2-large/circuits/factual_qa/hierarchical
|
| 79 |
-
15:04:59 | INFO | hub: restored 4 file(s) for gpt2-large/graphs/factual_qa
|
| 80 |
-
15:04:59 | INFO | hub: restored 2 file(s) for pythia-2.8b/circuits/factual_qa/hierarchical
|
| 81 |
-
15:05:00 | INFO | hub: restored 4 file(s) for pythia-2.8b/graphs/factual_qa
|
| 82 |
-
15:29:18 | INFO | transfer gpt2-large -> pythia-2.8b on factual_qa: tau = 0.030 (3145 aligned features)
|
| 83 |
-
15:29:18 | INFO | hub: restored 2 file(s) for gpt2-large/circuits/winogrande/hierarchical
|
| 84 |
-
15:29:20 | INFO | hub: restored 4 file(s) for gpt2-large/graphs/winogrande
|
| 85 |
-
15:29:20 | INFO | hub: restored 2 file(s) for pythia-2.8b/circuits/winogrande/hierarchical
|
| 86 |
-
15:29:21 | INFO | hub: restored 4 file(s) for pythia-2.8b/graphs/winogrande
|
| 87 |
-
15:29:21 | INFO | transfer gpt2-large -> pythia-2.8b on winogrande: tau = n/a (0 aligned features)
|
| 88 |
-
15:29:21 | INFO | hub: restored 2 file(s) for llama-7b/circuits/mod_arith/hierarchical
|
| 89 |
-
15:29:22 | INFO | hub: restored 4 file(s) for llama-7b/graphs/mod_arith
|
| 90 |
-
15:29:24 | INFO | transfer gpt2-large -> llama-7b on mod_arith: tau = 0.024 (871 aligned features)
|
| 91 |
-
15:29:24 | INFO | hub: restored 2 file(s) for llama-7b/circuits/factual_qa/hierarchical
|
| 92 |
-
15:29:25 | INFO | hub: restored 4 file(s) for llama-7b/graphs/factual_qa
|
| 93 |
-
15:53:50 | INFO | transfer gpt2-large -> llama-7b on factual_qa: tau = 0.010 (3001 aligned features)
|
| 94 |
-
15:53:50 | INFO | hub: restored 2 file(s) for llama-7b/circuits/winogrande/hierarchical
|
| 95 |
-
15:53:51 | INFO | hub: restored 4 file(s) for llama-7b/graphs/winogrande
|
| 96 |
-
15:53:51 | INFO | transfer gpt2-large -> llama-7b on winogrande: tau = 0.000 (1 aligned features)
|
| 97 |
-
15:53:51 | INFO | transfer pythia-2.8b -> llama-7b on mod_arith: tau = 0.013 (737 aligned features)
|
| 98 |
-
15:53:54 | INFO | transfer pythia-2.8b -> llama-7b on factual_qa: tau = 0.011 (2874 aligned features)
|
| 99 |
-
15:53:54 | INFO | transfer pythia-2.8b -> llama-7b on winogrande: tau = 0.000 (1 aligned features)
|
| 100 |
-
15:53:54 | INFO | transfer llama-7b -> llama-13b on mod_arith: circuits not available yet
|
| 101 |
-
15:53:54 | INFO | transfer llama-7b -> llama-13b on factual_qa: circuits not available yet
|
| 102 |
-
15:53:54 | INFO | transfer llama-7b -> llama-13b on winogrande: circuits not available yet
|
| 103 |
-
15:53:54 | INFO | transfer llama-7b -> llama-70b on mod_arith: circuits not available yet
|
| 104 |
-
15:53:54 | INFO | transfer llama-7b -> llama-70b on factual_qa: circuits not available yet
|
| 105 |
-
15:53:54 | INFO | transfer llama-7b -> llama-70b on winogrande: circuits not available yet
|
| 106 |
-
15:54:25 | INFO | === phase: complexity ===
|
| 107 |
-
15:54:25 | INFO | hub: restored 2 file(s) for analysis/complexity
|
| 108 |
-
15:54:25 | INFO | === phase: report ===
|
| 109 |
-
|
| 110 |
-
15:54:30 | INFO | hub: restored 325 result file(s) from earlier sessions
|
| 111 |
-
15:54:35 | INFO | report written to /home/marimo/hagdcodebase/runs/molab/report (72 circuits, 3 figures)
|
| 112 |
-
No files have been modified since last commit. Skipping to prevent empty commit.
|
| 113 |
-
15:54:37 | INFO | pipeline finished in 51.2 min; artifacts in /home/marimo/hagdcodebase/runs/molab
|
| 114 |
-
No files have been modified since last commit. Skipping to prevent empty commit.
|
| 115 |
-
16:34:16 | INFO | run molab | device cuda | models ['gpt2-small', 'gpt2-medium', 'gpt2-large', 'pythia-1.4b', 'pythia-2.8b', 'pythia-6.9b', 'llama-7b', 'llama-13b'] | label models ['pythia-70m', 'pythia-160m', 'gpt2-small']
|
| 116 |
-
16:34:16 | INFO | === phase: labels ===
|
| 117 |
-
16:34:16 | INFO | === phase: gnn ===
|
| 118 |
-
16:34:16 | INFO | === phase: circuits ===
|
| 119 |
-
16:34:16 | INFO | gpt2-small: all stages finished in an earlier session - skipping
|
| 120 |
-
16:34:16 | INFO | gpt2-medium: all stages finished in an earlier session - skipping
|
| 121 |
-
16:34:16 | INFO | gpt2-large: all stages finished in an earlier session - skipping
|
| 122 |
-
16:34:16 | INFO | pythia-1.4b: all stages finished in an earlier session - skipping
|
| 123 |
-
16:34:16 | INFO | pythia-2.8b: all stages finished in an earlier session - skipping
|
| 124 |
-
16:34:16 | INFO | pythia-6.9b: all stages finished in an earlier session - skipping
|
| 125 |
-
16:34:16 | INFO | llama-7b: all stages finished in an earlier session - skipping
|
| 126 |
-
16:34:17 | INFO | hub: 598 artifact file(s) of this run already on mudassiruddin/hagd
|
| 127 |
-
16:34:21 | WARNING | llama-13b: tokenizer of meta-llama/Llama-2-13b-hf unusable (You are trying to access a gated repo.
|
| 128 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h); using the identical tokenizer of NousResearch/Llama-2-7b-hf
|
| 129 |
-
|
| 130 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-hf.
|
| 131 |
-
403 Client Error. (Request ID: Root=1-
|
| 132 |
[transformers] Could not extract SentencePiece model from /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9fdd2ebd8db0901d37b98d3a53bbfaa4503/tokenizer.model using sentencepiece library due to
|
| 133 |
SentencePieceExtractor requires the SentencePiece library but it was not found in your environment. Check out the instructions on the
|
| 134 |
installation page of its repo: https://github.com/google/sentencepiece#installation and follow the ones
|
| 135 |
that match your environment. Please note that you may need to restart your runtime after installation.
|
| 136 |
. Falling back to TikToken extractor.
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
|
|
|
| 140 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h)
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
|
| 153 |
-
17:13:34 | INFO | adapt step 3500/12000 loss 0.7039 | mod_arith=0.969 parity=0.594 sorting=1.000
|
| 154 |
-
17:19:08 | INFO | adapt step 4000/12000 loss 0.0362 | mod_arith=1.000 parity=0.427 sorting=1.000
|
| 155 |
-
17:24:43 | INFO | adapt step 4500/12000 loss 0.0003 | mod_arith=1.000 parity=0.479 sorting=0.990
|
| 156 |
-
17:30:17 | INFO | adapt step 5000/12000 loss 0.6875 | mod_arith=1.000 parity=0.594 sorting=1.000
|
| 157 |
-
17:35:51 | INFO | adapt step 5500/12000 loss 0.0001 | mod_arith=1.000 parity=0.479 sorting=1.000
|
| 158 |
-
17:35:51 | WARNING | adapt: time budget of 60 min reached at step 5500
|
| 159 |
-
17:35:51 | INFO | adapt: kept step 5000 (validation accuracy mod_arith=1.000 parity=0.594 sorting=1.000)
|
| 160 |
-
17:36:44 | INFO | transcoder llama-13b step 1 | 0.00M/4.00M tokens | loss 4.4112 fvu 0.835 dead 0.000 | 1349 tok/s
|
| 161 |
-
|
| 162 |
-
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
17:37:41 | INFO | transcoder llama-13b step 50 | 0.08M/4.00M tokens | loss 4.2766 fvu 0.810 dead 0.000 | 1332 tok/s
|
| 169 |
-
17:38:24 | INFO | transcoder llama-13b step 100 | 0.15M/4.00M tokens | loss 11.7287 fvu 0.641 dead 0.000 | 1532 tok/s
|
| 170 |
-
17:39:05 | INFO | transcoder llama-13b step 150 | 0.23M/4.00M tokens | loss 3.2395 fvu 0.657 dead 0.000 | 1630 tok/s
|
| 171 |
-
17:39:47 | INFO | transcoder llama-13b step 200 | 0.31M/4.00M tokens | loss 3.1616 fvu 0.630 dead 0.000 | 1675 tok/s
|
| 172 |
-
17:40:31 | INFO | transcoder llama-13b step 250 | 0.38M/4.00M tokens | loss 2.8565 fvu 0.580 dead 0.000 | 1688 tok/s
|
| 173 |
-
17:41:13 | INFO | transcoder llama-13b step 300 | 0.46M/4.00M tokens | loss 4.1949 fvu 0.684 dead 0.000 | 1707 tok/s
|
| 174 |
-
17:41:54 | INFO | transcoder llama-13b step 350 | 0.54M/4.00M tokens | loss 2.7183 fvu 0.525 dead 0.000 | 1729 tok/s
|
| 175 |
-
17:42:38 | INFO | transcoder llama-13b step 400 | 0.61M/4.00M tokens | loss 2.8118 fvu 0.533 dead 0.000 | 1735 tok/s
|
| 176 |
-
17:43:20 | INFO | transcoder llama-13b step 450 | 0.69M/4.00M tokens | loss 3.0859 fvu 0.563 dead 0.000 | 1742 tok/s
|
| 177 |
-
17:44:06 | INFO | transcoder llama-13b step 500 | 0.77M/4.00M tokens | loss 2.7701 fvu 0.524 dead 0.000 | 1736 tok/s
|
| 178 |
-
17:44:51 | INFO | transcoder llama-13b step 550 | 0.84M/4.00M tokens | loss 4.1209 fvu 0.641 dead 0.000 | 1733 tok/s
|
| 179 |
-
17:45:33 | INFO | transcoder llama-13b step 600 | 0.92M/4.00M tokens | loss 2.4896 fvu 0.487 dead 0.000 | 1741 tok/s
|
| 180 |
-
17:46:16 | INFO | transcoder llama-13b step 650 | 1.00M/4.00M tokens | loss 2.7535 fvu 0.531 dead 0.000 | 1742 tok/s
|
| 181 |
-
17:47:01 | INFO | transcoder llama-13b step 700 | 1.08M/4.00M tokens | loss 3.8143 fvu 0.374 dead 0.000 | 1742 tok/s
|
| 182 |
-
17:47:44 | INFO | transcoder llama-13b step 750 | 1.15M/4.00M tokens | loss 2.9426 fvu 0.507 dead 0.000 | 1742 tok/s
|
| 183 |
-
17:48:32 | INFO | transcoder llama-13b step 800 | 1.23M/4.00M tokens | loss 2.4003 fvu 0.450 dead 0.000 | 1733 tok/s
|
| 184 |
-
17:49:18 | INFO | transcoder llama-13b step 850 | 1.31M/4.00M tokens | loss 2.4935 fvu 0.439 dead 0.000 | 1731 tok/s
|
| 185 |
-
17:50:02 | INFO | transcoder llama-13b step 900 | 1.38M/4.00M tokens | loss 2.4802 fvu 0.452 dead 0.000 | 1730 tok/s
|
| 186 |
-
17:50:49 | INFO | transcoder llama-13b step 950 | 1.46M/4.00M tokens | loss 2.5868 fvu 0.437 dead 0.000 | 1725 tok/s
|
| 187 |
-
17:51:36 | INFO | transcoder llama-13b step 1000 | 1.54M/4.00M tokens | loss 2.2252 fvu 0.406 dead 0.000 | 1720 tok/s
|
| 188 |
-
17:53:06 | INFO | transcoder llama-13b step 1050 | 1.61M/4.00M tokens | loss 2.4665 fvu 0.442 dead 0.000 | 1642 tok/s
|
| 189 |
-
17:53:54 | INFO | transcoder llama-13b step 1100 | 1.69M/4.00M tokens | loss 2.1427 fvu 0.395 dead 0.000 | 1639 tok/s
|
| 190 |
-
17:54:37 | INFO | transcoder llama-13b step 1150 | 1.77M/4.00M tokens | loss 3.7062 fvu 0.290 dead 0.000 | 1645 tok/s
|
| 191 |
-
17:55:26 | INFO | transcoder llama-13b step 1200 | 1.84M/4.00M tokens | loss 41.6795 fvu 3.210 dead 0.000 | 1642 tok/s
|
| 192 |
-
17:56:15 | INFO | transcoder llama-13b step 1250 | 1.92M/4.00M tokens | loss 2.2763 fvu 0.409 dead 0.000 | 1639 tok/s
|
| 193 |
-
17:57:02 | INFO | transcoder llama-13b step 1300 | 2.00M/4.00M tokens | loss 2.0604 fvu 0.388 dead 0.000 | 1638 tok/s
|
| 194 |
-
17:57:47 | INFO | transcoder llama-13b step 1350 | 2.07M/4.00M tokens | loss 2.1341 fvu 0.372 dead 0.000 | 1641 tok/s
|
| 195 |
-
17:58:35 | INFO | transcoder llama-13b step 1400 | 2.15M/4.00M tokens | loss 2.6424 fvu 0.231 dead 0.000 | 1640 tok/s
|
| 196 |
-
17:59:19 | INFO | transcoder llama-13b step 1450 | 2.23M/4.00M tokens | loss 9.9404 fvu 0.945 dead 0.000 | 1643 tok/s
|
| 197 |
-
18:00:06 | INFO | transcoder llama-13b step 1500 | 2.30M/4.00M tokens | loss 2.0333 fvu 0.369 dead 0.000 | 1643 tok/s
|
| 198 |
-
18:00:57 | INFO | transcoder llama-13b step 1550 | 2.38M/4.00M tokens | loss 2.3033 fvu 0.396 dead 0.000 | 1638 tok/s
|
| 199 |
-
18:01:46 | INFO | transcoder llama-13b step 1600 | 2.46M/4.00M tokens | loss 1.8544 fvu 0.347 dead 0.000 | 1635 tok/s
|
| 200 |
-
18:02:33 | INFO | transcoder llama-13b step 1650 | 2.53M/4.00M tokens | loss 2.1023 fvu 0.376 dead 0.000 | 1635 tok/s
|
| 201 |
-
18:03:21 | INFO | transcoder llama-13b step 1700 | 2.61M/4.00M tokens | loss 2.0798 fvu 0.380 dead 0.000 | 1634 tok/s
|
| 202 |
-
18:04:12 | INFO | transcoder llama-13b step 1750 | 2.69M/4.00M tokens | loss 2.1755 fvu 0.396 dead 0.000 | 1630 tok/s
|
| 203 |
-
18:05:01 | INFO | transcoder llama-13b step 1800 | 2.76M/4.00M tokens | loss 2.1218 fvu 0.374 dead 0.000 | 1628 tok/s
|
| 204 |
-
18:05:50 | INFO | transcoder llama-13b step 1850 | 2.84M/4.00M tokens | loss 2.2743 fvu 0.396 dead 0.000 | 1627 tok/s
|
| 205 |
-
18:06:39 | INFO | transcoder llama-13b step 1900 | 2.92M/4.00M tokens | loss 2.0507 fvu 0.361 dead 0.000 | 1625 tok/s
|
| 206 |
-
18:08:10 | INFO | transcoder llama-13b step 1950 | 3.00M/4.00M tokens | loss 2.0808 fvu 0.345 dead 0.000 | 1588 tok/s
|
| 207 |
-
18:08:57 | INFO | transcoder llama-13b step 2000 | 3.07M/4.00M tokens | loss 2.0297 fvu 0.222 dead 0.001 | 1589 tok/s
|
| 208 |
-
18:09:45 | INFO | transcoder llama-13b step 2050 | 3.15M/4.00M tokens | loss 1.9345 fvu 0.332 dead 0.001 | 1589 tok/s
|
| 209 |
-
18:10:36 | INFO | transcoder llama-13b step 2100 | 3.23M/4.00M tokens | loss 2.2596 fvu 0.245 dead 0.001 | 1587 tok/s
|
| 210 |
-
18:11:27 | INFO | transcoder llama-13b step 2150 | 3.30M/4.00M tokens | loss 1.8383 fvu 0.316 dead 0.001 | 1585 tok/s
|
| 211 |
-
18:12:13 | INFO | transcoder llama-13b step 2200 | 3.38M/4.00M tokens | loss 1.7920 fvu 0.321 dead 0.000 | 1587 tok/s
|
| 212 |
-
18:13:04 | INFO | transcoder llama-13b step 2250 | 3.46M/4.00M tokens | loss 2.2722 fvu 0.397 dead 0.000 | 1585 tok/s
|
| 213 |
-
18:13:56 | INFO | transcoder llama-13b step 2300 | 3.53M/4.00M tokens | loss 1.6774 fvu 0.217 dead 0.000 | 1583 tok/s
|
| 214 |
-
18:14:47 | INFO | transcoder llama-13b step 2350 | 3.61M/4.00M tokens | loss 2.1063 fvu 0.327 dead 0.001 | 1580 tok/s
|
| 215 |
-
18:15:37 | INFO | transcoder llama-13b step 2400 | 3.69M/4.00M tokens | loss 2.2716 fvu 0.375 dead 0.001 | 1580 tok/s
|
| 216 |
-
18:16:26 | INFO | transcoder llama-13b step 2450 | 3.76M/4.00M tokens | loss 2.1192 fvu 0.384 dead 0.001 | 1580 tok/s
|
| 217 |
-
18:17:15 | INFO | transcoder llama-13b step 2500 | 3.84M/4.00M tokens | loss 2.2466 fvu 0.259 dead 0.001 | 1579 tok/s
|
| 218 |
-
18:18:08 | INFO | transcoder llama-13b step 2550 | 3.92M/4.00M tokens | loss 2.0664 fvu 0.224 dead 0.001 | 1576 tok/s
|
| 219 |
-
18:19:00 | INFO | transcoder llama-13b step 2600 | 3.99M/4.00M tokens | loss 1.8383 fvu 0.316 dead 0.001 | 1575 tok/s
|
| 220 |
-
18:20:27 | INFO | transcoder llama-13b done: mean held-out FVU 0.355 over 5 layers (4001280 tokens)
|
| 221 |
-
18:25:10 | INFO | graph llama-13b/mod_arith: 2592 vertices (of 204800 features), 9915 edges | nodes 2.9s edges 222.6s
|
| 222 |
-
18:34:01 | INFO | circuit llama-13b/mod_arith/hierarchical: 849 nodes, 6489 edges, phi=0.945 (full acc 1.000), 686s
|
| 223 |
-
18:34:57 | INFO | circuit llama-13b/mod_arith/attribution_patching: 1020 nodes, 6015 edges, phi=0.926 (full acc 1.000), 240s
|
| 224 |
-
18:35:52 | INFO | circuit llama-13b/mod_arith/sparse_probing: 1020 nodes, 0 edges, phi=0.926 (full acc 1.000), 18s
|
| 225 |
-
18:35:54 | INFO | ACDC llama-13b/mod_arith: the 12-feature/layer pool itself scores -0.073 < theta; pruning cannot recover it
|
| 226 |
-
18:37:13 | INFO | ACDC llama-13b/mod_arith: 588 -> 0 edges, 4 vertices (ok, 81s)
|
| 227 |
-
18:37:54 | INFO | circuit llama-13b/mod_arith/acdc: 4 nodes, 0 edges, phi=0.000 (full acc 1.000), 84s
|
| 228 |
-
18:37:57 | INFO | llama-13b/parity: model accuracy 0.514 is at chance (0.500; needs 0.550) - no circuit search
|
| 229 |
-
18:46:40 | INFO | graph llama-13b/sorting: 2592 vertices (of 204800 features), 12865 edges | nodes 3.7s edges 486.5s
|
| 230 |
-
No files have been modified since last commit. Skipping to prevent empty commit.
|
| 231 |
-
18:59:42 | INFO | circuit llama-13b/sorting/hierarchical: 996 nodes, 10081 edges, phi=0.824 (full acc 1.000), 1162s
|
| 232 |
-
19:11:09 | INFO | graph llama-13b/winogrande: 2592 vertices (of 204800 features), 6345 edges | nodes 4.6s edges 680.0s
|
| 233 |
-
/usr/local/lib/python3.13/site-packages/sklearn/base.py:1403: ConvergenceWarning: Number of distinct clusters (132) found smaller than n_clusters (426). Possibly due to duplicate points in X.
|
| 234 |
-
return fit_method(estimator, *args, **kwargs)
|
|
|
|
| 1 |
No files have been modified since last commit. Skipping to prevent empty commit.
|
| 2 |
+
23:34:54 | INFO | run molab | device cuda | models ['gpt2-small', 'gpt2-medium', 'gpt2-large', 'pythia-1.4b', 'pythia-2.8b', 'pythia-6.9b', 'llama-7b', 'llama-13b'] | label models ['pythia-70m', 'pythia-160m', 'gpt2-small']
|
| 3 |
+
23:34:54 | INFO | === phase: labels ===
|
| 4 |
+
23:34:54 | INFO | hub: 632 artifact file(s) of this run already on mudassiruddin/hagd
|
| 5 |
+
23:34:54 | INFO | hub: restored 2 file(s) for pythia-70m/labels/mod_arith
|
| 6 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/parity
|
| 7 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/sorting
|
| 8 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/winogrande
|
| 9 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/hellaswag
|
| 10 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/factual_qa
|
| 11 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/ioi
|
| 12 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/greater_than
|
| 13 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/induction
|
| 14 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/mod_arith
|
| 15 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/parity
|
| 16 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/sorting
|
| 17 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/winogrande
|
| 18 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/hellaswag
|
| 19 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/factual_qa
|
| 20 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/ioi
|
| 21 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/greater_than
|
| 22 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/induction
|
| 23 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/mod_arith
|
| 24 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/parity
|
| 25 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/sorting
|
| 26 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/winogrande
|
| 27 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/hellaswag
|
| 28 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/factual_qa
|
| 29 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/ioi
|
| 30 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/greater_than
|
| 31 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/induction
|
| 32 |
+
23:35:00 | INFO | === phase: gnn ===
|
| 33 |
+
23:35:01 | INFO | hub: restored 3 file(s) for gnn/all
|
| 34 |
+
23:35:02 | INFO | hub: restored 3 file(s) for gnn/holdout_gpt2-small
|
| 35 |
+
23:35:02 | INFO | === phase: circuits ===
|
| 36 |
+
23:35:02 | INFO | hub: restored 1 file(s) for gpt2-small/complete
|
| 37 |
+
23:35:02 | INFO | gpt2-small: all stages finished in an earlier session - skipping
|
| 38 |
+
23:35:02 | INFO | hub: restored 1 file(s) for gpt2-medium/complete
|
| 39 |
+
23:35:02 | INFO | gpt2-medium: all stages finished in an earlier session - skipping
|
| 40 |
+
23:35:02 | INFO | hub: restored 1 file(s) for gpt2-large/complete
|
| 41 |
+
23:35:02 | INFO | gpt2-large: all stages finished in an earlier session - skipping
|
| 42 |
+
23:35:02 | INFO | hub: restored 1 file(s) for pythia-1.4b/complete
|
| 43 |
+
23:35:02 | INFO | pythia-1.4b: all stages finished in an earlier session - skipping
|
| 44 |
+
23:35:02 | INFO | hub: restored 1 file(s) for pythia-2.8b/complete
|
| 45 |
+
23:35:02 | INFO | pythia-2.8b: all stages finished in an earlier session - skipping
|
| 46 |
+
23:35:02 | INFO | hub: restored 1 file(s) for pythia-6.9b/complete
|
| 47 |
+
23:35:02 | INFO | pythia-6.9b: all stages finished in an earlier session - skipping
|
| 48 |
+
23:35:02 | INFO | hub: restored 1 file(s) for llama-7b/complete
|
| 49 |
+
23:35:02 | INFO | llama-7b: all stages finished in an earlier session - skipping
|
| 50 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/hierarchical
|
| 51 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/parity/hierarchical
|
| 52 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/sorting/hierarchical
|
| 53 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/winogrande/hierarchical
|
| 54 |
+
23:35:09 | WARNING | llama-13b: tokenizer of meta-llama/Llama-2-13b-hf unusable (You are trying to access a gated repo.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h); using the identical tokenizer of NousResearch/Llama-2-7b-hf
|
| 56 |
+
23:35:09 | WARNING | could not load meta-llama/Llama-2-13b-hf: You are trying to access a gated repo.
|
| 57 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-hf.
|
| 58 |
+
403 Client Error. (Request ID: Root=1-6afe362d-7aa5a0042d94c40d1a034c3c;c2fd2
|
| 59 |
[transformers] Could not extract SentencePiece model from /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9fdd2ebd8db0901d37b98d3a53bbfaa4503/tokenizer.model using sentencepiece library due to
|
| 60 |
SentencePieceExtractor requires the SentencePiece library but it was not found in your environment. Check out the instructions on the
|
| 61 |
installation page of its repo: https://github.com/google/sentencepiece#installation and follow the ones
|
| 62 |
that match your environment. Please note that you may need to restart your runtime after installation.
|
| 63 |
. Falling back to TikToken extractor.
|
| 64 |
+
23:35:10 | WARNING | llama-13b: tokenizer of NousResearch/Llama-2-13b-hf unusable (Error parsing line b'\x0e' in /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9f); using the identical tokenizer of NousResearch/Llama-2-7b-hf
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
23:36:08 | WARNING | llama-13b: loaded fallback repo NousResearch/Llama-2-13b-hf (primary meta-llama/Llama-2-13b-hf unavailable: You are trying to access a gated repo.
|
| 68 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h)
|
| 69 |
+
23:36:08 | INFO | loaded llama-13b (NousResearch/Llama-2-13b-hf): 40 layers, d=5120, dtype=torch.bfloat16, device=cuda:0
|
| 70 |
+
23:36:08 | INFO | LoRA: wrapped 280 linear maps (rank 32, 125.17M trainable parameters)
|
| 71 |
+
23:36:15 | INFO | hub: restored 3 file(s) for llama-13b/adapter
|
| 72 |
+
23:36:15 | INFO | llama-13b: adapter restored (val acc {'step': 5000, 'mod_arith': 1.0, 'parity': 0.59375, 'sorting': 1.0})
|
| 73 |
+
23:36:37 | INFO | hub: restored 5 file(s) for llama-13b/transcoder
|
| 74 |
+
23:36:49 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/attribution_patching
|
| 75 |
+
23:36:49 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/sparse_probing
|
| 76 |
+
23:36:49 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/acdc
|
| 77 |
+
|
| 78 |
+
|
| 79 |
+
|
| 80 |
+
01:20:22 | INFO | graph llama-13b/hellaswag: 2592 vertices (of 204800 features), 4821 edges | nodes 29.6s edges 6162.7s
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
runs/molab/logs/pipeline.log
CHANGED
|
@@ -1,204 +1,69 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
Traceback (most recent call last):
|
| 55 |
-
File "/home/marimo/hagdcodebase/src/hagd/pipeline.py", line 545, in _open
|
| 56 |
-
ctx = ModelContext(self, key, tasks)
|
| 57 |
-
File "/home/marimo/hagdcodebase/src/hagd/pipeline.py", line 58, in __init__
|
| 58 |
-
self.lm = LM.load(model_spec(pipe.cfg, key), pipe.device, self.cfg.get("quantization"))
|
| 59 |
-
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 60 |
-
File "/home/marimo/hagdcodebase/src/hagd/models.py", line 119, in load
|
| 61 |
-
raise RuntimeError(f"failed to load {spec.key}: {last_err}")
|
| 62 |
-
RuntimeError: failed to load llama-13b: Error parsing line b'\x0e' in /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9fdd2ebd8db0901d37b98d3a53bbfaa4503/tokenizer.model
|
| 63 |
-
|
| 64 |
-
15:04:53 | INFO | === phase: transfer ===
|
| 65 |
-
15:04:53 | INFO | hub: restored 2 file(s) for gpt2-large/circuits/mod_arith/hierarchical
|
| 66 |
-
15:04:55 | INFO | hub: restored 4 file(s) for gpt2-large/graphs/mod_arith
|
| 67 |
-
15:04:55 | INFO | hub: restored 2 file(s) for pythia-2.8b/circuits/mod_arith/hierarchical
|
| 68 |
-
15:04:56 | INFO | hub: restored 4 file(s) for pythia-2.8b/graphs/mod_arith
|
| 69 |
-
15:04:58 | INFO | transfer gpt2-large -> pythia-2.8b on mod_arith: tau = 0.057 (767 aligned features)
|
| 70 |
-
15:04:58 | INFO | hub: restored 2 file(s) for gpt2-large/circuits/factual_qa/hierarchical
|
| 71 |
-
15:04:59 | INFO | hub: restored 4 file(s) for gpt2-large/graphs/factual_qa
|
| 72 |
-
15:04:59 | INFO | hub: restored 2 file(s) for pythia-2.8b/circuits/factual_qa/hierarchical
|
| 73 |
-
15:05:00 | INFO | hub: restored 4 file(s) for pythia-2.8b/graphs/factual_qa
|
| 74 |
-
15:29:18 | INFO | transfer gpt2-large -> pythia-2.8b on factual_qa: tau = 0.030 (3145 aligned features)
|
| 75 |
-
15:29:18 | INFO | hub: restored 2 file(s) for gpt2-large/circuits/winogrande/hierarchical
|
| 76 |
-
15:29:20 | INFO | hub: restored 4 file(s) for gpt2-large/graphs/winogrande
|
| 77 |
-
15:29:20 | INFO | hub: restored 2 file(s) for pythia-2.8b/circuits/winogrande/hierarchical
|
| 78 |
-
15:29:21 | INFO | hub: restored 4 file(s) for pythia-2.8b/graphs/winogrande
|
| 79 |
-
15:29:21 | INFO | transfer gpt2-large -> pythia-2.8b on winogrande: tau = n/a (0 aligned features)
|
| 80 |
-
15:29:21 | INFO | hub: restored 2 file(s) for llama-7b/circuits/mod_arith/hierarchical
|
| 81 |
-
15:29:22 | INFO | hub: restored 4 file(s) for llama-7b/graphs/mod_arith
|
| 82 |
-
15:29:24 | INFO | transfer gpt2-large -> llama-7b on mod_arith: tau = 0.024 (871 aligned features)
|
| 83 |
-
15:29:24 | INFO | hub: restored 2 file(s) for llama-7b/circuits/factual_qa/hierarchical
|
| 84 |
-
15:29:25 | INFO | hub: restored 4 file(s) for llama-7b/graphs/factual_qa
|
| 85 |
-
15:53:50 | INFO | transfer gpt2-large -> llama-7b on factual_qa: tau = 0.010 (3001 aligned features)
|
| 86 |
-
15:53:50 | INFO | hub: restored 2 file(s) for llama-7b/circuits/winogrande/hierarchical
|
| 87 |
-
15:53:51 | INFO | hub: restored 4 file(s) for llama-7b/graphs/winogrande
|
| 88 |
-
15:53:51 | INFO | transfer gpt2-large -> llama-7b on winogrande: tau = 0.000 (1 aligned features)
|
| 89 |
-
15:53:51 | INFO | transfer pythia-2.8b -> llama-7b on mod_arith: tau = 0.013 (737 aligned features)
|
| 90 |
-
15:53:54 | INFO | transfer pythia-2.8b -> llama-7b on factual_qa: tau = 0.011 (2874 aligned features)
|
| 91 |
-
15:53:54 | INFO | transfer pythia-2.8b -> llama-7b on winogrande: tau = 0.000 (1 aligned features)
|
| 92 |
-
15:53:54 | INFO | transfer llama-7b -> llama-13b on mod_arith: circuits not available yet
|
| 93 |
-
15:53:54 | INFO | transfer llama-7b -> llama-13b on factual_qa: circuits not available yet
|
| 94 |
-
15:53:54 | INFO | transfer llama-7b -> llama-13b on winogrande: circuits not available yet
|
| 95 |
-
15:53:54 | INFO | transfer llama-7b -> llama-70b on mod_arith: circuits not available yet
|
| 96 |
-
15:53:54 | INFO | transfer llama-7b -> llama-70b on factual_qa: circuits not available yet
|
| 97 |
-
15:53:54 | INFO | transfer llama-7b -> llama-70b on winogrande: circuits not available yet
|
| 98 |
-
15:54:25 | INFO | === phase: complexity ===
|
| 99 |
-
15:54:25 | INFO | hub: restored 2 file(s) for analysis/complexity
|
| 100 |
-
15:54:25 | INFO | === phase: report ===
|
| 101 |
-
15:54:30 | INFO | hub: restored 325 result file(s) from earlier sessions
|
| 102 |
-
15:54:35 | INFO | report written to /home/marimo/hagdcodebase/runs/molab/report (72 circuits, 3 figures)
|
| 103 |
-
15:54:37 | INFO | pipeline finished in 51.2 min; artifacts in /home/marimo/hagdcodebase/runs/molab
|
| 104 |
-
16:34:16 | INFO | run molab | device cuda | models ['gpt2-small', 'gpt2-medium', 'gpt2-large', 'pythia-1.4b', 'pythia-2.8b', 'pythia-6.9b', 'llama-7b', 'llama-13b'] | label models ['pythia-70m', 'pythia-160m', 'gpt2-small']
|
| 105 |
-
16:34:16 | INFO | === phase: labels ===
|
| 106 |
-
16:34:16 | INFO | === phase: gnn ===
|
| 107 |
-
16:34:16 | INFO | === phase: circuits ===
|
| 108 |
-
16:34:16 | INFO | gpt2-small: all stages finished in an earlier session - skipping
|
| 109 |
-
16:34:16 | INFO | gpt2-medium: all stages finished in an earlier session - skipping
|
| 110 |
-
16:34:16 | INFO | gpt2-large: all stages finished in an earlier session - skipping
|
| 111 |
-
16:34:16 | INFO | pythia-1.4b: all stages finished in an earlier session - skipping
|
| 112 |
-
16:34:16 | INFO | pythia-2.8b: all stages finished in an earlier session - skipping
|
| 113 |
-
16:34:16 | INFO | pythia-6.9b: all stages finished in an earlier session - skipping
|
| 114 |
-
16:34:16 | INFO | llama-7b: all stages finished in an earlier session - skipping
|
| 115 |
-
16:34:17 | INFO | hub: 598 artifact file(s) of this run already on mudassiruddin/hagd
|
| 116 |
-
16:34:21 | WARNING | llama-13b: tokenizer of meta-llama/Llama-2-13b-hf unusable (You are trying to access a gated repo.
|
| 117 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h); using the identical tokenizer of NousResearch/Llama-2-7b-hf
|
| 118 |
-
|
| 119 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-hf.
|
| 120 |
-
403 Client Error. (Request ID: Root=1-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h)
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
17:13:34 | INFO | adapt step 3500/12000 loss 0.7039 | mod_arith=0.969 parity=0.594 sorting=1.000
|
| 134 |
-
17:19:08 | INFO | adapt step 4000/12000 loss 0.0362 | mod_arith=1.000 parity=0.427 sorting=1.000
|
| 135 |
-
17:24:43 | INFO | adapt step 4500/12000 loss 0.0003 | mod_arith=1.000 parity=0.479 sorting=0.990
|
| 136 |
-
17:30:17 | INFO | adapt step 5000/12000 loss 0.6875 | mod_arith=1.000 parity=0.594 sorting=1.000
|
| 137 |
-
17:35:51 | INFO | adapt step 5500/12000 loss 0.0001 | mod_arith=1.000 parity=0.479 sorting=1.000
|
| 138 |
-
17:35:51 | WARNING | adapt: time budget of 60 min reached at step 5500
|
| 139 |
-
17:35:51 | INFO | adapt: kept step 5000 (validation accuracy mod_arith=1.000 parity=0.594 sorting=1.000)
|
| 140 |
-
17:36:44 | INFO | transcoder llama-13b step 1 | 0.00M/4.00M tokens | loss 4.4112 fvu 0.835 dead 0.000 | 1349 tok/s
|
| 141 |
-
17:37:41 | INFO | transcoder llama-13b step 50 | 0.08M/4.00M tokens | loss 4.2766 fvu 0.810 dead 0.000 | 1332 tok/s
|
| 142 |
-
17:38:24 | INFO | transcoder llama-13b step 100 | 0.15M/4.00M tokens | loss 11.7287 fvu 0.641 dead 0.000 | 1532 tok/s
|
| 143 |
-
17:39:05 | INFO | transcoder llama-13b step 150 | 0.23M/4.00M tokens | loss 3.2395 fvu 0.657 dead 0.000 | 1630 tok/s
|
| 144 |
-
17:39:47 | INFO | transcoder llama-13b step 200 | 0.31M/4.00M tokens | loss 3.1616 fvu 0.630 dead 0.000 | 1675 tok/s
|
| 145 |
-
17:40:31 | INFO | transcoder llama-13b step 250 | 0.38M/4.00M tokens | loss 2.8565 fvu 0.580 dead 0.000 | 1688 tok/s
|
| 146 |
-
17:41:13 | INFO | transcoder llama-13b step 300 | 0.46M/4.00M tokens | loss 4.1949 fvu 0.684 dead 0.000 | 1707 tok/s
|
| 147 |
-
17:41:54 | INFO | transcoder llama-13b step 350 | 0.54M/4.00M tokens | loss 2.7183 fvu 0.525 dead 0.000 | 1729 tok/s
|
| 148 |
-
17:42:38 | INFO | transcoder llama-13b step 400 | 0.61M/4.00M tokens | loss 2.8118 fvu 0.533 dead 0.000 | 1735 tok/s
|
| 149 |
-
17:43:20 | INFO | transcoder llama-13b step 450 | 0.69M/4.00M tokens | loss 3.0859 fvu 0.563 dead 0.000 | 1742 tok/s
|
| 150 |
-
17:44:06 | INFO | transcoder llama-13b step 500 | 0.77M/4.00M tokens | loss 2.7701 fvu 0.524 dead 0.000 | 1736 tok/s
|
| 151 |
-
17:44:51 | INFO | transcoder llama-13b step 550 | 0.84M/4.00M tokens | loss 4.1209 fvu 0.641 dead 0.000 | 1733 tok/s
|
| 152 |
-
17:45:33 | INFO | transcoder llama-13b step 600 | 0.92M/4.00M tokens | loss 2.4896 fvu 0.487 dead 0.000 | 1741 tok/s
|
| 153 |
-
17:46:16 | INFO | transcoder llama-13b step 650 | 1.00M/4.00M tokens | loss 2.7535 fvu 0.531 dead 0.000 | 1742 tok/s
|
| 154 |
-
17:47:01 | INFO | transcoder llama-13b step 700 | 1.08M/4.00M tokens | loss 3.8143 fvu 0.374 dead 0.000 | 1742 tok/s
|
| 155 |
-
17:47:44 | INFO | transcoder llama-13b step 750 | 1.15M/4.00M tokens | loss 2.9426 fvu 0.507 dead 0.000 | 1742 tok/s
|
| 156 |
-
17:48:32 | INFO | transcoder llama-13b step 800 | 1.23M/4.00M tokens | loss 2.4003 fvu 0.450 dead 0.000 | 1733 tok/s
|
| 157 |
-
17:49:18 | INFO | transcoder llama-13b step 850 | 1.31M/4.00M tokens | loss 2.4935 fvu 0.439 dead 0.000 | 1731 tok/s
|
| 158 |
-
17:50:02 | INFO | transcoder llama-13b step 900 | 1.38M/4.00M tokens | loss 2.4802 fvu 0.452 dead 0.000 | 1730 tok/s
|
| 159 |
-
17:50:49 | INFO | transcoder llama-13b step 950 | 1.46M/4.00M tokens | loss 2.5868 fvu 0.437 dead 0.000 | 1725 tok/s
|
| 160 |
-
17:51:36 | INFO | transcoder llama-13b step 1000 | 1.54M/4.00M tokens | loss 2.2252 fvu 0.406 dead 0.000 | 1720 tok/s
|
| 161 |
-
17:53:06 | INFO | transcoder llama-13b step 1050 | 1.61M/4.00M tokens | loss 2.4665 fvu 0.442 dead 0.000 | 1642 tok/s
|
| 162 |
-
17:53:54 | INFO | transcoder llama-13b step 1100 | 1.69M/4.00M tokens | loss 2.1427 fvu 0.395 dead 0.000 | 1639 tok/s
|
| 163 |
-
17:54:37 | INFO | transcoder llama-13b step 1150 | 1.77M/4.00M tokens | loss 3.7062 fvu 0.290 dead 0.000 | 1645 tok/s
|
| 164 |
-
17:55:26 | INFO | transcoder llama-13b step 1200 | 1.84M/4.00M tokens | loss 41.6795 fvu 3.210 dead 0.000 | 1642 tok/s
|
| 165 |
-
17:56:15 | INFO | transcoder llama-13b step 1250 | 1.92M/4.00M tokens | loss 2.2763 fvu 0.409 dead 0.000 | 1639 tok/s
|
| 166 |
-
17:57:02 | INFO | transcoder llama-13b step 1300 | 2.00M/4.00M tokens | loss 2.0604 fvu 0.388 dead 0.000 | 1638 tok/s
|
| 167 |
-
17:57:47 | INFO | transcoder llama-13b step 1350 | 2.07M/4.00M tokens | loss 2.1341 fvu 0.372 dead 0.000 | 1641 tok/s
|
| 168 |
-
17:58:35 | INFO | transcoder llama-13b step 1400 | 2.15M/4.00M tokens | loss 2.6424 fvu 0.231 dead 0.000 | 1640 tok/s
|
| 169 |
-
17:59:19 | INFO | transcoder llama-13b step 1450 | 2.23M/4.00M tokens | loss 9.9404 fvu 0.945 dead 0.000 | 1643 tok/s
|
| 170 |
-
18:00:06 | INFO | transcoder llama-13b step 1500 | 2.30M/4.00M tokens | loss 2.0333 fvu 0.369 dead 0.000 | 1643 tok/s
|
| 171 |
-
18:00:57 | INFO | transcoder llama-13b step 1550 | 2.38M/4.00M tokens | loss 2.3033 fvu 0.396 dead 0.000 | 1638 tok/s
|
| 172 |
-
18:01:46 | INFO | transcoder llama-13b step 1600 | 2.46M/4.00M tokens | loss 1.8544 fvu 0.347 dead 0.000 | 1635 tok/s
|
| 173 |
-
18:02:33 | INFO | transcoder llama-13b step 1650 | 2.53M/4.00M tokens | loss 2.1023 fvu 0.376 dead 0.000 | 1635 tok/s
|
| 174 |
-
18:03:21 | INFO | transcoder llama-13b step 1700 | 2.61M/4.00M tokens | loss 2.0798 fvu 0.380 dead 0.000 | 1634 tok/s
|
| 175 |
-
18:04:12 | INFO | transcoder llama-13b step 1750 | 2.69M/4.00M tokens | loss 2.1755 fvu 0.396 dead 0.000 | 1630 tok/s
|
| 176 |
-
18:05:01 | INFO | transcoder llama-13b step 1800 | 2.76M/4.00M tokens | loss 2.1218 fvu 0.374 dead 0.000 | 1628 tok/s
|
| 177 |
-
18:05:50 | INFO | transcoder llama-13b step 1850 | 2.84M/4.00M tokens | loss 2.2743 fvu 0.396 dead 0.000 | 1627 tok/s
|
| 178 |
-
18:06:39 | INFO | transcoder llama-13b step 1900 | 2.92M/4.00M tokens | loss 2.0507 fvu 0.361 dead 0.000 | 1625 tok/s
|
| 179 |
-
18:08:10 | INFO | transcoder llama-13b step 1950 | 3.00M/4.00M tokens | loss 2.0808 fvu 0.345 dead 0.000 | 1588 tok/s
|
| 180 |
-
18:08:57 | INFO | transcoder llama-13b step 2000 | 3.07M/4.00M tokens | loss 2.0297 fvu 0.222 dead 0.001 | 1589 tok/s
|
| 181 |
-
18:09:45 | INFO | transcoder llama-13b step 2050 | 3.15M/4.00M tokens | loss 1.9345 fvu 0.332 dead 0.001 | 1589 tok/s
|
| 182 |
-
18:10:36 | INFO | transcoder llama-13b step 2100 | 3.23M/4.00M tokens | loss 2.2596 fvu 0.245 dead 0.001 | 1587 tok/s
|
| 183 |
-
18:11:27 | INFO | transcoder llama-13b step 2150 | 3.30M/4.00M tokens | loss 1.8383 fvu 0.316 dead 0.001 | 1585 tok/s
|
| 184 |
-
18:12:13 | INFO | transcoder llama-13b step 2200 | 3.38M/4.00M tokens | loss 1.7920 fvu 0.321 dead 0.000 | 1587 tok/s
|
| 185 |
-
18:13:04 | INFO | transcoder llama-13b step 2250 | 3.46M/4.00M tokens | loss 2.2722 fvu 0.397 dead 0.000 | 1585 tok/s
|
| 186 |
-
18:13:56 | INFO | transcoder llama-13b step 2300 | 3.53M/4.00M tokens | loss 1.6774 fvu 0.217 dead 0.000 | 1583 tok/s
|
| 187 |
-
18:14:47 | INFO | transcoder llama-13b step 2350 | 3.61M/4.00M tokens | loss 2.1063 fvu 0.327 dead 0.001 | 1580 tok/s
|
| 188 |
-
18:15:37 | INFO | transcoder llama-13b step 2400 | 3.69M/4.00M tokens | loss 2.2716 fvu 0.375 dead 0.001 | 1580 tok/s
|
| 189 |
-
18:16:26 | INFO | transcoder llama-13b step 2450 | 3.76M/4.00M tokens | loss 2.1192 fvu 0.384 dead 0.001 | 1580 tok/s
|
| 190 |
-
18:17:15 | INFO | transcoder llama-13b step 2500 | 3.84M/4.00M tokens | loss 2.2466 fvu 0.259 dead 0.001 | 1579 tok/s
|
| 191 |
-
18:18:08 | INFO | transcoder llama-13b step 2550 | 3.92M/4.00M tokens | loss 2.0664 fvu 0.224 dead 0.001 | 1576 tok/s
|
| 192 |
-
18:19:00 | INFO | transcoder llama-13b step 2600 | 3.99M/4.00M tokens | loss 1.8383 fvu 0.316 dead 0.001 | 1575 tok/s
|
| 193 |
-
18:20:27 | INFO | transcoder llama-13b done: mean held-out FVU 0.355 over 5 layers (4001280 tokens)
|
| 194 |
-
18:25:10 | INFO | graph llama-13b/mod_arith: 2592 vertices (of 204800 features), 9915 edges | nodes 2.9s edges 222.6s
|
| 195 |
-
18:34:01 | INFO | circuit llama-13b/mod_arith/hierarchical: 849 nodes, 6489 edges, phi=0.945 (full acc 1.000), 686s
|
| 196 |
-
18:34:57 | INFO | circuit llama-13b/mod_arith/attribution_patching: 1020 nodes, 6015 edges, phi=0.926 (full acc 1.000), 240s
|
| 197 |
-
18:35:52 | INFO | circuit llama-13b/mod_arith/sparse_probing: 1020 nodes, 0 edges, phi=0.926 (full acc 1.000), 18s
|
| 198 |
-
18:35:54 | INFO | ACDC llama-13b/mod_arith: the 12-feature/layer pool itself scores -0.073 < theta; pruning cannot recover it
|
| 199 |
-
18:37:13 | INFO | ACDC llama-13b/mod_arith: 588 -> 0 edges, 4 vertices (ok, 81s)
|
| 200 |
-
18:37:54 | INFO | circuit llama-13b/mod_arith/acdc: 4 nodes, 0 edges, phi=0.000 (full acc 1.000), 84s
|
| 201 |
-
18:37:57 | INFO | llama-13b/parity: model accuracy 0.514 is at chance (0.500; needs 0.550) - no circuit search
|
| 202 |
-
18:46:40 | INFO | graph llama-13b/sorting: 2592 vertices (of 204800 features), 12865 edges | nodes 3.7s edges 486.5s
|
| 203 |
-
18:59:42 | INFO | circuit llama-13b/sorting/hierarchical: 996 nodes, 10081 edges, phi=0.824 (full acc 1.000), 1162s
|
| 204 |
-
19:11:09 | INFO | graph llama-13b/winogrande: 2592 vertices (of 204800 features), 6345 edges | nodes 4.6s edges 680.0s
|
|
|
|
| 1 |
+
23:34:54 | INFO | run molab | device cuda | models ['gpt2-small', 'gpt2-medium', 'gpt2-large', 'pythia-1.4b', 'pythia-2.8b', 'pythia-6.9b', 'llama-7b', 'llama-13b'] | label models ['pythia-70m', 'pythia-160m', 'gpt2-small']
|
| 2 |
+
23:34:54 | INFO | === phase: labels ===
|
| 3 |
+
23:34:54 | INFO | hub: 632 artifact file(s) of this run already on mudassiruddin/hagd
|
| 4 |
+
23:34:54 | INFO | hub: restored 2 file(s) for pythia-70m/labels/mod_arith
|
| 5 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/parity
|
| 6 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/sorting
|
| 7 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/winogrande
|
| 8 |
+
23:34:55 | INFO | hub: restored 2 file(s) for pythia-70m/labels/hellaswag
|
| 9 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/factual_qa
|
| 10 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/ioi
|
| 11 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/greater_than
|
| 12 |
+
23:34:56 | INFO | hub: restored 2 file(s) for pythia-70m/labels/induction
|
| 13 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/mod_arith
|
| 14 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/parity
|
| 15 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/sorting
|
| 16 |
+
23:34:57 | INFO | hub: restored 2 file(s) for pythia-160m/labels/winogrande
|
| 17 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/hellaswag
|
| 18 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/factual_qa
|
| 19 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/ioi
|
| 20 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/greater_than
|
| 21 |
+
23:34:58 | INFO | hub: restored 2 file(s) for pythia-160m/labels/induction
|
| 22 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/mod_arith
|
| 23 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/parity
|
| 24 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/sorting
|
| 25 |
+
23:34:59 | INFO | hub: restored 2 file(s) for gpt2-small/labels/winogrande
|
| 26 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/hellaswag
|
| 27 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/factual_qa
|
| 28 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/ioi
|
| 29 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/greater_than
|
| 30 |
+
23:35:00 | INFO | hub: restored 2 file(s) for gpt2-small/labels/induction
|
| 31 |
+
23:35:00 | INFO | === phase: gnn ===
|
| 32 |
+
23:35:01 | INFO | hub: restored 3 file(s) for gnn/all
|
| 33 |
+
23:35:02 | INFO | hub: restored 3 file(s) for gnn/holdout_gpt2-small
|
| 34 |
+
23:35:02 | INFO | === phase: circuits ===
|
| 35 |
+
23:35:02 | INFO | hub: restored 1 file(s) for gpt2-small/complete
|
| 36 |
+
23:35:02 | INFO | gpt2-small: all stages finished in an earlier session - skipping
|
| 37 |
+
23:35:02 | INFO | hub: restored 1 file(s) for gpt2-medium/complete
|
| 38 |
+
23:35:02 | INFO | gpt2-medium: all stages finished in an earlier session - skipping
|
| 39 |
+
23:35:02 | INFO | hub: restored 1 file(s) for gpt2-large/complete
|
| 40 |
+
23:35:02 | INFO | gpt2-large: all stages finished in an earlier session - skipping
|
| 41 |
+
23:35:02 | INFO | hub: restored 1 file(s) for pythia-1.4b/complete
|
| 42 |
+
23:35:02 | INFO | pythia-1.4b: all stages finished in an earlier session - skipping
|
| 43 |
+
23:35:02 | INFO | hub: restored 1 file(s) for pythia-2.8b/complete
|
| 44 |
+
23:35:02 | INFO | pythia-2.8b: all stages finished in an earlier session - skipping
|
| 45 |
+
23:35:02 | INFO | hub: restored 1 file(s) for pythia-6.9b/complete
|
| 46 |
+
23:35:02 | INFO | pythia-6.9b: all stages finished in an earlier session - skipping
|
| 47 |
+
23:35:02 | INFO | hub: restored 1 file(s) for llama-7b/complete
|
| 48 |
+
23:35:02 | INFO | llama-7b: all stages finished in an earlier session - skipping
|
| 49 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/hierarchical
|
| 50 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/parity/hierarchical
|
| 51 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/sorting/hierarchical
|
| 52 |
+
23:35:03 | INFO | hub: restored 2 file(s) for llama-13b/circuits/winogrande/hierarchical
|
| 53 |
+
23:35:09 | WARNING | llama-13b: tokenizer of meta-llama/Llama-2-13b-hf unusable (You are trying to access a gated repo.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h); using the identical tokenizer of NousResearch/Llama-2-7b-hf
|
| 55 |
+
23:35:09 | WARNING | could not load meta-llama/Llama-2-13b-hf: You are trying to access a gated repo.
|
| 56 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-hf.
|
| 57 |
+
403 Client Error. (Request ID: Root=1-6afe362d-7aa5a0042d94c40d1a034c3c;c2fd2
|
| 58 |
+
23:35:10 | WARNING | llama-13b: tokenizer of NousResearch/Llama-2-13b-hf unusable (Error parsing line b'\x0e' in /home/marimo/.cache/huggingface/hub/models--NousResearch--Llama-2-13b-hf/snapshots/4b54e9f); using the identical tokenizer of NousResearch/Llama-2-7b-hf
|
| 59 |
+
23:36:08 | WARNING | llama-13b: loaded fallback repo NousResearch/Llama-2-13b-hf (primary meta-llama/Llama-2-13b-hf unavailable: You are trying to access a gated repo.
|
| 60 |
Make sure to have access to it at https://huggingface.co/meta-llama/Llama-2-13b-h)
|
| 61 |
+
23:36:08 | INFO | loaded llama-13b (NousResearch/Llama-2-13b-hf): 40 layers, d=5120, dtype=torch.bfloat16, device=cuda:0
|
| 62 |
+
23:36:08 | INFO | LoRA: wrapped 280 linear maps (rank 32, 125.17M trainable parameters)
|
| 63 |
+
23:36:15 | INFO | hub: restored 3 file(s) for llama-13b/adapter
|
| 64 |
+
23:36:15 | INFO | llama-13b: adapter restored (val acc {'step': 5000, 'mod_arith': 1.0, 'parity': 0.59375, 'sorting': 1.0})
|
| 65 |
+
23:36:37 | INFO | hub: restored 5 file(s) for llama-13b/transcoder
|
| 66 |
+
23:36:49 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/attribution_patching
|
| 67 |
+
23:36:49 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/sparse_probing
|
| 68 |
+
23:36:49 | INFO | hub: restored 2 file(s) for llama-13b/circuits/mod_arith/acdc
|
| 69 |
+
01:20:22 | INFO | graph llama-13b/hellaswag: 2592 vertices (of 204800 features), 4821 edges | nodes 29.6s edges 6162.7s
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|