| ================================================================================ | |
| STARTING UNIFIED 4-MODEL HARDENED HashedNets BENCHMARK ON GPU SERVER | |
| Running models: ['distilbert', 'roberta', 'gpt2'] | |
| ================================================================================ | |
| ============================== RUNNING MODEL: distilbert ============================== | |
| 2026-09-24 19:19:00,710 [INFO] QwenQualitySweep: GPU VRAM: 6.89 GB free / 8.00 GB total on NVIDIA GeForce RTX 4070 Laptop GPU | |
| 2026-09-24 19:19:00,710 [INFO] QwenQualitySweep: Configured execution sizing: parallel=4, batch_size=8 (explicit_parallel=False, explicit_batch=True) | |
| 2026-09-24 19:19:00,733 [INFO] QwenQualitySweep: Running 1 compressor modules via isolated subprocesses (workers=4) | |
| [ISOLATED] Launching hashednets (timeout=180s) -> logs\isolated\hashednets.log | |
| [ISOLATED] hashednets exited with code 0 in 37.6s | |
| 2026-09-24 19:19:40,541 [INFO] QwenQualitySweep: Running 1 attention modules in-memory sequentially | |
| 2026-09-24 19:19:40,780 [INFO] QwenQualitySweep: Configured dataset 'sst2' with num_labels=2 | |
| Loading weights: 0%| | 0/104 [00:00<?, ?it/s] | |
| Loading weights: 100%|##########| 104/104 [00:00<00:00, 2978.44it/s] | |
| 2026-09-24 19:19:41,329 [INFO] QwenQualitySweep: model=models\distilbert_sst2 qwen_compat=False hidden=768 heads=12 head_dim=? layers=6 | |
| 2026-09-24 19:19:41,330 [INFO] QwenQualitySweep: detected model family=bert adapter={'family': 'bert', 'hidden_size': 768, 'num_heads': 12, 'num_key_value_heads': 12, 'q_proj_out': 768, 'kv_proj_out': 768, 'kv_repeat': 1, 'head_dim': 64, 'num_hidden_layers': 6, 'causal': False, 'mask_rank': 2, 'position_embedding': 'absolute', 'supports_gqa': False, 'supports_delta_net': False, 'supports_moe': False} | |
| 2026-09-24 19:19:41,330 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:19:41,330 [INFO] QwenQualitySweep: ACTIVE SWEEP CONFIGURATION: | |
| 2026-09-24 19:19:41,330 [INFO] QwenQualitySweep: model: models\distilbert_sst2 | |
| 2026-09-24 19:19:41,330 [INFO] QwenQualitySweep: family: bert | |
| 2026-09-24 19:19:41,330 [INFO] QwenQualitySweep: dataset: sst2 | |
| 2026-09-24 19:19:41,331 [INFO] QwenQualitySweep: batch_size: 8 | |
| 2026-09-24 19:19:41,331 [INFO] QwenQualitySweep: recovery_steps: 50 | |
| 2026-09-24 19:19:41,331 [INFO] QwenQualitySweep: optimizer: adamw (muon_lr=2.0e-02, adam_lr=2.0e-04, weight_decay=1.0e-02) | |
| 2026-09-24 19:19:41,331 [INFO] QwenQualitySweep: precision: fp16 | |
| 2026-09-24 19:19:41,331 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:19:41,331 [WARNING] QwenQualitySweep: Specified parent_checkpoint 'results\unified_asvd\distilbert\finetuned_model.pt' does not exist; attempting automatic fallback search. | |
| 2026-09-24 19:19:41,336 [INFO] QwenQualitySweep: Loading cached fine-tuned checkpoint: results\unified_hashednets\distilbert\finetuned_model.pt | |
| 2026-09-24 19:19:41,344 [WARNING] QwenQualitySweep: Missing keys during load: ['distilbert.embeddings.word_embeddings.weight', 'distilbert.embeddings.position_embeddings.weight', 'distilbert.embeddings.LayerNorm.weight', 'distilbert.embeddings.LayerNorm.bias', 'distilbert.transformer.layer.0.attention.q_lin.weight', 'distilbert.transformer.layer.0.attention.q_lin.bias', 'distilbert.transformer.layer.0.attention.k_lin.weight', 'distilbert.transformer.layer.0.attention.k_lin.bias', 'distilbert.transformer.layer.0.attention.v_lin.weight', 'distilbert.transformer.layer.0.attention.v_lin.bias'] | |
| 2026-09-24 19:19:41,450 [INFO] QwenQualitySweep: Loading pre-cached batches from results\unified_hashednets\distilbert\batch_cache_sst2 (skipping tokenization) | |
| 2026-09-24 19:19:41,757 [INFO] QwenQualitySweep: Prepared 109 eval batches and 50 train batches directly on GPU. | |
| 2026-09-24 19:19:44,893 [INFO] QwenQualitySweep: Baseline model benchmark accuracy on validation set: 90.71% | |
| 2026-09-24 19:19:45,009 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:19:45,009 [INFO] QwenQualitySweep: STATISTICAL BENCHMARK THRESHOLDS (Validation N=872): | |
| 2026-09-24 19:19:45,009 [INFO] QwenQualitySweep: Validation Majority Count: 444 / 872 | |
| 2026-09-24 19:19:45,009 [INFO] QwenQualitySweep: True Majority Baseline: 50.92% | |
| 2026-09-24 19:19:45,009 [INFO] QwenQualitySweep: Standard Error (SE): 1.693% | |
| 2026-09-24 19:19:45,010 [INFO] QwenQualitySweep: Statistical Margin (3*SE): 5.08% | |
| 2026-09-24 19:19:45,010 [INFO] QwenQualitySweep: PASS Threshold (Min Acc): 56.00% | |
| 2026-09-24 19:19:45,010 [INFO] QwenQualitySweep: Threshold Source: dynamic: majority=50.92% (444/872), SE=1.693%, 3*SE=5.080% | |
| 2026-09-24 19:19:45,010 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:19:45,021 [INFO] QwenQualitySweep: Initializing ModelStateSnapshot for in-memory module swapping... | |
| 2026-09-24 19:19:45,120 [INFO] QwenQualitySweep: >>> STARTING MODULE [1/1]: full (recovery_steps=50, batch_size=8, optimizer=adamw) | |
| 2026-09-24 19:19:48,330 [INFO] QwenQualitySweep: Exported comprehensive profiling metrics to results\unified_hashednets\distilbert\distilbert_sst2_full_full_metrics.json | |
| 2026-09-24 19:19:48,483 [INFO] QwenQualitySweep: [1/1] full -> PASS 90.71% (F1: 0.9099, Peak VRAM: 380.26 MB) | |
| 2026-09-24 19:19:48,483 [INFO] QwenQualitySweep: [Export Deferred] Checkpoint preserved in results\unified_hashednets\distilbert\checkpoints\full. Automated export will be handled post-sweep by orchestrator. | |
| 2026-09-24 19:19:48,484 [INFO] QwenQualitySweep: [1/1] full Timing: eval_time=3.36s | |
| 2026-09-24 19:19:48,484 [INFO] QwenQualitySweep: Restoring model state from snapshot for next module... | |
| 2026-09-24 19:19:48,807 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:19:48,808 [INFO] QwenQualitySweep: BENCHMARK QUALITY SWEEP SUMMARY TABLE (distilbert_sst2 on sst2) | |
| 2026-09-24 19:19:48,816 [INFO] QwenQualitySweep: | |
| Module Status Accuracy Delta Acc F1 Compressed Layers Compression Rate Peak VRAM Inference Time (s) Latency (ms/samp) | |
| hashednets (efficient-attn) PASS_NO_COMPRESSION 90.71% +0.00% 0.9099 0 1.00x 291.9 MB 3.59 N/A | |
| full PASS 90.71% 0.00% (base) 0.9099 0 1.00x 380.3 MB 2.45 2.81 | |
| 2026-09-24 19:19:48,817 [INFO] QwenQualitySweep: ================================================================================ | |
| >>> FINISHED MODEL: distilbert in 75.4s (Exit: 0, Status: PASS_NO_COMPRESSION, Compressed Layers: 0, Comp Rate: 1.00x) | |
| ============================== RUNNING MODEL: roberta ============================== | |
| 2026-09-24 19:20:17,372 [INFO] QwenQualitySweep: GPU VRAM: 5.03 GB free / 8.00 GB total on NVIDIA GeForce RTX 4070 Laptop GPU | |
| 2026-09-24 19:20:17,373 [INFO] QwenQualitySweep: Configured execution sizing: parallel=4, batch_size=8 (explicit_parallel=False, explicit_batch=True) | |
| 2026-09-24 19:20:17,395 [INFO] QwenQualitySweep: Running 1 compressor modules via isolated subprocesses (workers=4) | |
| [ISOLATED] Launching hashednets (timeout=180s) -> logs\isolated\hashednets.log | |
| [ISOLATED] hashednets exited with code 0 in 59.0s | |
| 2026-09-24 19:21:18,619 [INFO] QwenQualitySweep: Running 1 attention modules in-memory sequentially | |
| 2026-09-24 19:21:19,185 [INFO] QwenQualitySweep: Configured dataset 'sst2' with num_labels=2 | |
| Loading weights: 0%| | 0/201 [00:00<?, ?it/s] | |
| Loading weights: 100%|##########| 201/201 [00:00<00:00, 2872.34it/s] | |
| [1mRobertaForSequenceClassification LOAD REPORT[0m from: models\roberta_sst2 | |
| Key | Status | | | |
| ----------------------------+------------+--+- | |
| roberta.pooler.dense.bias | UNEXPECTED | | | |
| roberta.pooler.dense.weight | UNEXPECTED | | | |
| Notes: | |
| - UNEXPECTED: can be ignored when loading from different task/architecture; not ok if you expect identical arch. | |
| 2026-09-24 19:21:23,507 [INFO] QwenQualitySweep: model=models\roberta_sst2 qwen_compat=False hidden=768 heads=12 head_dim=? layers=12 | |
| 2026-09-24 19:21:23,507 [INFO] QwenQualitySweep: detected model family=roberta adapter={'family': 'roberta', 'hidden_size': 768, 'num_heads': 12, 'num_key_value_heads': 12, 'q_proj_out': 768, 'kv_proj_out': 768, 'kv_repeat': 1, 'head_dim': 64, 'num_hidden_layers': 12, 'causal': False, 'mask_rank': 2, 'position_embedding': 'absolute', 'supports_gqa': False, 'supports_delta_net': False, 'supports_moe': False} | |
| 2026-09-24 19:21:23,507 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:21:23,507 [INFO] QwenQualitySweep: ACTIVE SWEEP CONFIGURATION: | |
| 2026-09-24 19:21:23,508 [INFO] QwenQualitySweep: model: models\roberta_sst2 | |
| 2026-09-24 19:21:23,508 [INFO] QwenQualitySweep: family: roberta | |
| 2026-09-24 19:21:23,508 [INFO] QwenQualitySweep: dataset: sst2 | |
| 2026-09-24 19:21:23,508 [INFO] QwenQualitySweep: batch_size: 8 | |
| 2026-09-24 19:21:23,508 [INFO] QwenQualitySweep: recovery_steps: 50 | |
| 2026-09-24 19:21:23,508 [INFO] QwenQualitySweep: optimizer: adamw (muon_lr=2.0e-02, adam_lr=2.0e-04, weight_decay=1.0e-02) | |
| 2026-09-24 19:21:23,509 [INFO] QwenQualitySweep: precision: fp16 | |
| 2026-09-24 19:21:23,509 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:21:23,509 [WARNING] QwenQualitySweep: Specified parent_checkpoint 'results\roberta_sst2_lora_baseline\finetuned_model.pt' does not exist; attempting automatic fallback search. | |
| 2026-09-24 19:21:23,519 [INFO] QwenQualitySweep: Loading cached fine-tuned checkpoint: results\unified_hashednets\roberta\finetuned_model.pt | |
| 2026-09-24 19:21:23,535 [WARNING] QwenQualitySweep: Missing keys during load: ['roberta.embeddings.word_embeddings.weight', 'roberta.embeddings.token_type_embeddings.weight', 'roberta.embeddings.LayerNorm.weight', 'roberta.embeddings.LayerNorm.bias', 'roberta.embeddings.position_embeddings.weight', 'roberta.encoder.layer.0.attention.self.query.weight', 'roberta.encoder.layer.0.attention.self.query.bias', 'roberta.encoder.layer.0.attention.self.key.weight', 'roberta.encoder.layer.0.attention.self.key.bias', 'roberta.encoder.layer.0.attention.self.value.weight'] | |
| 2026-09-24 19:21:24,001 [INFO] QwenQualitySweep: Loading pre-cached batches from results\unified_hashednets\roberta\batch_cache_sst2 (skipping tokenization) | |
| 2026-09-24 19:21:28,908 [INFO] QwenQualitySweep: Prepared 109 eval batches and 50 train batches directly on GPU. | |
| 2026-09-24 19:21:44,396 [INFO] QwenQualitySweep: Baseline model benchmark accuracy on validation set: 94.38% | |
| 2026-09-24 19:21:44,669 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:21:44,669 [INFO] QwenQualitySweep: STATISTICAL BENCHMARK THRESHOLDS (Validation N=872): | |
| 2026-09-24 19:21:44,670 [INFO] QwenQualitySweep: Validation Majority Count: 444 / 872 | |
| 2026-09-24 19:21:44,670 [INFO] QwenQualitySweep: True Majority Baseline: 50.92% | |
| 2026-09-24 19:21:44,670 [INFO] QwenQualitySweep: Standard Error (SE): 1.693% | |
| 2026-09-24 19:21:44,670 [INFO] QwenQualitySweep: Statistical Margin (3*SE): 5.08% | |
| 2026-09-24 19:21:44,670 [INFO] QwenQualitySweep: PASS Threshold (Min Acc): 56.00% | |
| 2026-09-24 19:21:44,670 [INFO] QwenQualitySweep: Threshold Source: dynamic: majority=50.92% (444/872), SE=1.693%, 3*SE=5.080% | |
| 2026-09-24 19:21:44,670 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:21:44,701 [INFO] QwenQualitySweep: Initializing ModelStateSnapshot for in-memory module swapping... | |
| 2026-09-24 19:21:45,645 [INFO] QwenQualitySweep: >>> STARTING MODULE [1/1]: full (recovery_steps=50, batch_size=8, optimizer=adamw) | |
| 2026-09-24 19:21:57,515 [INFO] QwenQualitySweep: Exported comprehensive profiling metrics to results\unified_hashednets\roberta\roberta_sst2_full_full_metrics.json | |
| 2026-09-24 19:21:58,690 [INFO] QwenQualitySweep: [1/1] full -> PASS 94.38% (F1: 0.9445, Peak VRAM: 703.34 MB) | |
| 2026-09-24 19:21:58,691 [INFO] QwenQualitySweep: [Export Deferred] Checkpoint preserved in results\unified_hashednets\roberta\checkpoints\full. Automated export will be handled post-sweep by orchestrator. | |
| 2026-09-24 19:21:58,691 [INFO] QwenQualitySweep: [1/1] full Timing: eval_time=13.05s | |
| 2026-09-24 19:21:58,691 [INFO] QwenQualitySweep: Restoring model state from snapshot for next module... | |
| 2026-09-24 19:21:59,058 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:21:59,058 [INFO] QwenQualitySweep: BENCHMARK QUALITY SWEEP SUMMARY TABLE (roberta_sst2 on sst2) | |
| 2026-09-24 19:21:59,082 [INFO] QwenQualitySweep: | |
| Module Status Accuracy Delta Acc F1 Compressed Layers Compression Rate Peak VRAM Inference Time (s) Latency (ms/samp) | |
| hashednets (efficient-attn) PASS_NO_COMPRESSION 94.38% +0.00% 0.9445 0 1.00x 511.4 MB 23.88 N/A | |
| full PASS 94.38% 0.00% (base) 0.9445 0 1.00x 703.3 MB 7.77 8.91 | |
| 2026-09-24 19:21:59,083 [INFO] QwenQualitySweep: ================================================================================ | |
| >>> FINISHED MODEL: roberta in 130.9s (Exit: 0, Status: PASS_NO_COMPRESSION, Compressed Layers: 0, Comp Rate: 1.00x) | |
| ============================== RUNNING MODEL: gpt2 ============================== | |
| 2026-09-24 19:22:26,449 [INFO] QwenQualitySweep: GPU VRAM: 6.89 GB free / 8.00 GB total on NVIDIA GeForce RTX 4070 Laptop GPU | |
| 2026-09-24 19:22:26,450 [INFO] QwenQualitySweep: Configured execution sizing: parallel=4, batch_size=8 (explicit_parallel=False, explicit_batch=True) | |
| 2026-09-24 19:22:26,477 [INFO] QwenQualitySweep: Running 1 compressor modules via isolated subprocesses (workers=4) | |
| [ISOLATED] Launching hashednets (timeout=180s) -> logs\isolated\hashednets.log | |
| [ISOLATED] hashednets exited with code 0 in 146.2s | |
| 2026-09-24 19:24:55,076 [INFO] QwenQualitySweep: Running 1 attention modules in-memory sequentially | |
| 2026-09-24 19:24:55,578 [INFO] QwenQualitySweep: Configured dataset 'sst2' with num_labels=2 | |
| Loading weights: 0%| | 0/149 [00:00<?, ?it/s] | |
| Loading weights: 100%|##########| 149/149 [00:00<00:00, 3760.85it/s] | |
| 2026-09-24 19:25:29,585 [INFO] QwenQualitySweep: model=models\gpt2_sst2 qwen_compat=False hidden=768 heads=12 head_dim=? layers=12 | |
| 2026-09-24 19:25:29,590 [INFO] QwenQualitySweep: detected model family=gpt adapter={'family': 'gpt', 'hidden_size': 768, 'num_heads': 12, 'num_key_value_heads': 12, 'q_proj_out': 768, 'kv_proj_out': 768, 'kv_repeat': 1, 'head_dim': 64, 'num_hidden_layers': 12, 'causal': True, 'mask_rank': 4, 'position_embedding': 'absolute', 'supports_gqa': False, 'supports_delta_net': False, 'supports_moe': False} | |
| 2026-09-24 19:25:29,591 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:25:29,591 [INFO] QwenQualitySweep: ACTIVE SWEEP CONFIGURATION: | |
| 2026-09-24 19:25:29,592 [INFO] QwenQualitySweep: model: models\gpt2_sst2 | |
| 2026-09-24 19:25:29,643 [INFO] QwenQualitySweep: family: gpt | |
| 2026-09-24 19:25:29,643 [INFO] QwenQualitySweep: dataset: sst2 | |
| 2026-09-24 19:25:29,643 [INFO] QwenQualitySweep: batch_size: 8 | |
| 2026-09-24 19:25:29,643 [INFO] QwenQualitySweep: recovery_steps: 50 | |
| 2026-09-24 19:25:29,643 [INFO] QwenQualitySweep: optimizer: adamw (muon_lr=2.0e-02, adam_lr=2.0e-04, weight_decay=1.0e-02) | |
| 2026-09-24 19:25:29,643 [INFO] QwenQualitySweep: precision: fp16 | |
| 2026-09-24 19:25:29,643 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:25:29,645 [WARNING] QwenQualitySweep: Specified parent_checkpoint 'results\gpt2_sst2_lora_baseline\finetuned_model.pt' does not exist; attempting automatic fallback search. | |
| 2026-09-24 19:25:29,694 [INFO] QwenQualitySweep: Loading cached fine-tuned checkpoint: results\unified_hashednets\gpt2\finetuned_model.pt | |
| 2026-09-24 19:25:29,704 [WARNING] QwenQualitySweep: Missing keys during load: ['transformer.wte.weight', 'transformer.wpe.weight', 'transformer.h.0.ln_1.weight', 'transformer.h.0.ln_1.bias', 'transformer.h.0.attn.c_attn.weight', 'transformer.h.0.attn.c_attn.bias', 'transformer.h.0.attn.c_proj.weight', 'transformer.h.0.attn.c_proj.bias', 'transformer.h.0.ln_2.weight', 'transformer.h.0.ln_2.bias'] | |
| 2026-09-24 19:25:30,456 [INFO] QwenQualitySweep: Loading pre-cached batches from results\unified_hashednets\gpt2\batch_cache_sst2 (skipping tokenization) | |
| 2026-09-24 19:25:36,930 [INFO] QwenQualitySweep: Prepared 109 eval batches and 50 train batches directly on GPU. | |
| 2026-09-24 19:26:25,435 [INFO] QwenQualitySweep: Baseline model benchmark accuracy on validation set: 89.33% | |
| 2026-09-24 19:26:30,298 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:26:30,298 [INFO] QwenQualitySweep: STATISTICAL BENCHMARK THRESHOLDS (Validation N=872): | |
| 2026-09-24 19:26:30,298 [INFO] QwenQualitySweep: Validation Majority Count: 444 / 872 | |
| 2026-09-24 19:26:30,299 [INFO] QwenQualitySweep: True Majority Baseline: 50.92% | |
| 2026-09-24 19:26:30,299 [INFO] QwenQualitySweep: Standard Error (SE): 1.693% | |
| 2026-09-24 19:26:30,299 [INFO] QwenQualitySweep: Statistical Margin (3*SE): 5.08% | |
| 2026-09-24 19:26:30,299 [INFO] QwenQualitySweep: PASS Threshold (Min Acc): 56.00% | |
| 2026-09-24 19:26:30,299 [INFO] QwenQualitySweep: Threshold Source: dynamic: majority=50.92% (444/872), SE=1.693%, 3*SE=5.080% | |
| 2026-09-24 19:26:30,299 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:26:30,338 [INFO] QwenQualitySweep: Initializing ModelStateSnapshot for in-memory module swapping... | |
| 2026-09-24 19:26:31,793 [INFO] QwenQualitySweep: >>> STARTING MODULE [1/1]: full (recovery_steps=50, batch_size=8, optimizer=adamw) | |
| 2026-09-24 19:26:41,021 [INFO] QwenQualitySweep: Exported comprehensive profiling metrics to results\unified_hashednets\gpt2\gpt2_sst2_full_full_metrics.json | |
| 2026-09-24 19:26:41,550 [INFO] QwenQualitySweep: [1/1] full -> PASS 89.33% (F1: 0.8942, Peak VRAM: 784.19 MB) | |
| 2026-09-24 19:26:41,552 [INFO] QwenQualitySweep: [Export Deferred] Checkpoint preserved in results\unified_hashednets\gpt2\checkpoints\full. Automated export will be handled post-sweep by orchestrator. | |
| 2026-09-24 19:26:41,553 [INFO] QwenQualitySweep: [1/1] full Timing: eval_time=9.76s | |
| 2026-09-24 19:26:41,553 [INFO] QwenQualitySweep: Restoring model state from snapshot for next module... | |
| 2026-09-24 19:26:41,944 [INFO] QwenQualitySweep: ================================================================================ | |
| 2026-09-24 19:26:41,944 [INFO] QwenQualitySweep: BENCHMARK QUALITY SWEEP SUMMARY TABLE (gpt2_sst2 on sst2) | |
| 2026-09-24 19:26:41,968 [INFO] QwenQualitySweep: | |
| Module Status Accuracy Delta Acc F1 Compressed Layers Compression Rate Peak VRAM Inference Time (s) Latency (ms/samp) | |
| hashednets (efficient-attn) PASS_NO_COMPRESSION 89.33% +0.00% 0.8942 0 1.00x 383.6 MB 48.17 N/A | |
| full PASS 89.33% 0.00% (base) 0.8942 0 1.00x 784.2 MB 5.67 6.50 | |
| 2026-09-24 19:26:41,968 [INFO] QwenQualitySweep: ================================================================================ | |
| >>> FINISHED MODEL: gpt2 in 283.1s (Exit: 0, Status: PASS_NO_COMPRESSION, Compressed Layers: 0, Comp Rate: 1.00x) | |
| ================================================================================ | |
| 4-MODEL BENCHMARK RESULTS SUMMARY (HashedNets on GPU) | |
| ================================================================================ | |
| Model Exit Code Baseline Acc HashedNets Acc Delta Acc F1 Status Compressed Layers Comp Rate Peak VRAM Time (s) | |
| distilbert 0 90.71% 90.71% +0.00% 0.9099 PASS_NO_COMPRESSION 0 1.00x 291.9 MB 75.4 | |
| roberta 0 94.38% 94.38% +0.00% 0.9445 PASS_NO_COMPRESSION 0 1.00x 511.4 MB 130.9 | |
| gpt2 0 89.33% 89.33% +0.00% 0.8942 PASS_NO_COMPRESSION 0 1.00x 383.6 MB 283.1 | |
| ================================================================================ | |
| Saved consolidated summaries to: | |
| - results\unified_hashednets_4models_summary.csv | |
| - results\unified_hashednets_4models_summary.json | |
| Generated and saved comparative visualization charts to results\unified_baseline | |
| ================================================================================ | |
| STAGE 2: QUALITY-GATED EXPORT TO HUGGING FACE & GITHUB (HashedNets) | |
| ================================================================================ | |
| [Quality Gate: REJECTED] distilbert did not meet criteria (status=PASS_NO_COMPRESSION, acc=90.71% < 56.0%). | |
| Pruning local checkpoint artifacts in results\unified_hashednets\distilbert\checkpoints\hashednets | |
| [Quality Gate: REJECTED] roberta did not meet criteria (status=PASS_NO_COMPRESSION, acc=94.38% < 56.0%). | |
| Pruning local checkpoint artifacts in results\unified_hashednets\roberta\checkpoints\hashednets | |
| [Quality Gate: REJECTED] gpt2 did not meet criteria (status=PASS_NO_COMPRESSION, acc=89.33% < 56.0%). | |
| Pruning local checkpoint artifacts in results\unified_hashednets\gpt2\checkpoints\hashednets | |
| [+] Updated documentation benchmark table in Docs\content\compressions\hashed_net_compressor.md | |
| Committing results directly to main... | |
| Results committed to main. | |
| ================================================================================ | |
| ALL TASKS COMPLETED! | |
| Passed Models: 0 / 3 | |
| Uploaded Repositories: 0 | |
| ================================================================================ | |
Xet Storage Details
- Size:
- 22.6 kB
- Xet hash:
- 270ed608d2dc1ec1c16fb9dfa20226bb242afeef01cd5b46c16d86a7386ef220
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.