Upload evidence/build-A0.log with huggingface_hub
Browse files- evidence/build-A0.log +70 -0
evidence/build-A0.log
ADDED
|
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
[transformers] [1mGemma4UnifiedModel LOAD REPORT[0m from: /content/ltxwork/encoder-hf
|
| 3 |
+
Key | Status | |
|
| 4 |
+
-------------------------------------------------------+------------+--+-
|
| 5 |
+
tokenizer_json | UNEXPECTED | |
|
| 6 |
+
text_embedding_projection.audio_aggregate_embed.weight | UNEXPECTED | |
|
| 7 |
+
text_embedding_projection.audio_aggregate_embed.bias | UNEXPECTED | |
|
| 8 |
+
hf_asset__processor_config.json | UNEXPECTED | |
|
| 9 |
+
hf_asset__tokenizer_config.json | UNEXPECTED | |
|
| 10 |
+
hf_asset__chat_template.jinja | UNEXPECTED | |
|
| 11 |
+
text_embedding_projection.video_aggregate_embed.bias | UNEXPECTED | |
|
| 12 |
+
text_embedding_projection.video_aggregate_embed.weight | UNEXPECTED | |
|
| 13 |
+
hf_asset__generation_config.json | UNEXPECTED | |
|
| 14 |
+
|
| 15 |
+
Notes:
|
| 16 |
+
- UNEXPECTED: can be ignored when loading from different task/architecture; not ok if you expect identical arch.
|
| 17 |
+
captured contexts for 33 distinct lengths
|
| 18 |
+
calibration: 146 batches, 166028 tokens
|
| 19 |
+
wrote 22 non-decoder tensors
|
| 20 |
+
layer 0 [slid] 7 quantized, worst diagonal output relL2 0.1169, worst ||Q||/||W|| 302.654, dead columns 3
|
| 21 |
+
layer 1 [slid] 7 quantized, worst diagonal output relL2 0.1263, worst ||Q||/||W|| 2.614, dead columns 2
|
| 22 |
+
layer 2 [slid] 7 quantized, worst diagonal output relL2 0.1109, worst ||Q||/||W|| 1.011, dead columns 0
|
| 23 |
+
layer 3 [slid] 7 quantized, worst diagonal output relL2 0.1163, worst ||Q||/||W|| 1.010, dead columns 0
|
| 24 |
+
layer 4 [slid] 7 quantized, worst diagonal output relL2 0.1146, worst ||Q||/||W|| 1.006, dead columns 0
|
| 25 |
+
layer 5 [full] 5 quantized, worst diagonal output relL2 0.1146, worst ||Q||/||W|| 1.007, dead columns 0
|
| 26 |
+
layer 6 [slid] 7 quantized, worst diagonal output relL2 0.1178, worst ||Q||/||W|| 1.009, dead columns 0
|
| 27 |
+
layer 7 [slid] 7 quantized, worst diagonal output relL2 0.1171, worst ||Q||/||W|| 1.007, dead columns 0
|
| 28 |
+
layer 8 [slid] 7 quantized, worst diagonal output relL2 0.1096, worst ||Q||/||W|| 1.014, dead columns 0
|
| 29 |
+
layer 9 [slid] 7 quantized, worst diagonal output relL2 0.1086, worst ||Q||/||W|| 1.010, dead columns 0
|
| 30 |
+
layer 10 [slid] 7 quantized, worst diagonal output relL2 0.1105, worst ||Q||/||W|| 1.081, dead columns 0
|
| 31 |
+
layer 11 [full] 5 quantized, worst diagonal output relL2 0.1107, worst ||Q||/||W|| 1.010, dead columns 0
|
| 32 |
+
layer 12 [slid] 7 quantized, worst diagonal output relL2 0.1180, worst ||Q||/||W|| 1.145, dead columns 0
|
| 33 |
+
layer 13 [slid] 7 quantized, worst diagonal output relL2 0.1176, worst ||Q||/||W|| 1.380, dead columns 0
|
| 34 |
+
layer 14 [slid] 7 quantized, worst diagonal output relL2 0.1176, worst ||Q||/||W|| 1.068, dead columns 0
|
| 35 |
+
layer 15 [slid] 7 quantized, worst diagonal output relL2 0.1111, worst ||Q||/||W|| 1.026, dead columns 0
|
| 36 |
+
layer 16 [slid] 7 quantized, worst diagonal output relL2 0.1075, worst ||Q||/||W|| 1.039, dead columns 0
|
| 37 |
+
layer 17 [full] 5 quantized, worst diagonal output relL2 0.1102, worst ||Q||/||W|| 1.019, dead columns 0
|
| 38 |
+
layer 18 [slid] 7 quantized, worst diagonal output relL2 0.1132, worst ||Q||/||W|| 1.010, dead columns 0
|
| 39 |
+
layer 19 [slid] 7 quantized, worst diagonal output relL2 0.1112, worst ||Q||/||W|| 1.006, dead columns 0
|
| 40 |
+
layer 20 [slid] 7 quantized, worst diagonal output relL2 0.1154, worst ||Q||/||W|| 1.008, dead columns 0
|
| 41 |
+
layer 21 [slid] 7 quantized, worst diagonal output relL2 0.1162, worst ||Q||/||W|| 1.012, dead columns 0
|
| 42 |
+
layer 22 [slid] 7 quantized, worst diagonal output relL2 0.1171, worst ||Q||/||W|| 1.007, dead columns 0
|
| 43 |
+
layer 23 [full] 5 quantized, worst diagonal output relL2 0.1156, worst ||Q||/||W|| 1.011, dead columns 0
|
| 44 |
+
layer 24 [slid] 7 quantized, worst diagonal output relL2 0.1159, worst ||Q||/||W|| 1.006, dead columns 0
|
| 45 |
+
layer 25 [slid] 7 quantized, worst diagonal output relL2 0.1123, worst ||Q||/||W|| 1.005, dead columns 0
|
| 46 |
+
layer 26 [slid] 7 quantized, worst diagonal output relL2 0.1136, worst ||Q||/||W|| 1.005, dead columns 0
|
| 47 |
+
layer 27 [slid] 7 quantized, worst diagonal output relL2 0.1118, worst ||Q||/||W|| 1.006, dead columns 0
|
| 48 |
+
layer 28 [slid] 7 quantized, worst diagonal output relL2 0.1170, worst ||Q||/||W|| 1.006, dead columns 0
|
| 49 |
+
layer 29 [full] 5 quantized, worst diagonal output relL2 0.1113, worst ||Q||/||W|| 1.006, dead columns 0
|
| 50 |
+
layer 30 [slid] 7 quantized, worst diagonal output relL2 0.1100, worst ||Q||/||W|| 1.005, dead columns 0
|
| 51 |
+
layer 31 [slid] 7 quantized, worst diagonal output relL2 0.1033, worst ||Q||/||W|| 1.005, dead columns 0
|
| 52 |
+
layer 32 [slid] 7 quantized, worst diagonal output relL2 0.1091, worst ||Q||/||W|| 1.006, dead columns 0
|
| 53 |
+
layer 33 [slid] 7 quantized, worst diagonal output relL2 0.0997, worst ||Q||/||W|| 1.010, dead columns 0
|
| 54 |
+
layer 34 [slid] 7 quantized, worst diagonal output relL2 0.1210, worst ||Q||/||W|| 1.004, dead columns 0
|
| 55 |
+
layer 35 [full] 5 quantized, worst diagonal output relL2 0.1170, worst ||Q||/||W|| 1.007, dead columns 0
|
| 56 |
+
layer 36 [slid] 7 quantized, worst diagonal output relL2 0.1169, worst ||Q||/||W|| 1.005, dead columns 0
|
| 57 |
+
layer 37 [slid] 7 quantized, worst diagonal output relL2 0.1159, worst ||Q||/||W|| 1.005, dead columns 0
|
| 58 |
+
layer 38 [slid] 7 quantized, worst diagonal output relL2 0.1141, worst ||Q||/||W|| 1.006, dead columns 0
|
| 59 |
+
layer 39 [slid] 7 quantized, worst diagonal output relL2 0.1148, worst ||Q||/||W|| 1.005, dead columns 0
|
| 60 |
+
layer 40 [slid] 7 quantized, worst diagonal output relL2 0.1139, worst ||Q||/||W|| 1.004, dead columns 0
|
| 61 |
+
layer 41 [full] 5 quantized, worst diagonal output relL2 0.1128, worst ||Q||/||W|| 1.004, dead columns 0
|
| 62 |
+
layer 42 [slid] 7 quantized, worst diagonal output relL2 0.1144, worst ||Q||/||W|| 1.003, dead columns 0
|
| 63 |
+
layer 43 [slid] 7 quantized, worst diagonal output relL2 0.1132, worst ||Q||/||W|| 1.003, dead columns 0
|
| 64 |
+
layer 44 [slid] 7 quantized, worst diagonal output relL2 0.1092, worst ||Q||/||W|| 1.004, dead columns 0
|
| 65 |
+
layer 45 [slid] 7 quantized, worst diagonal output relL2 0.1045, worst ||Q||/||W|| 1.005, dead columns 0
|
| 66 |
+
layer 46 [slid] 7 quantized, worst diagonal output relL2 0.0996, worst ||Q||/||W|| 1.008, dead columns 0
|
| 67 |
+
layer 47 [full] 5 quantized, worst diagonal output relL2 0.0870, worst ||Q||/||W|| 1.015, dead columns 0
|
| 68 |
+
dead-guard legacy eps 1e-10 awq-floor 1e-05 restore-gain-max 64 damping mean awq-group none
|
| 69 |
+
worst ||Q||/||W|| 302.654 | columns zeroed 5 | max restore gain 469.56 | channels exact-zero 5 near-dead 967 | worst damping escalation 1x
|
| 70 |
+
wrote /content/builds/A0.packed.safetensors (8.464 GB packed, 19945148 zero-sign collapses)
|