Upload pre_redeploy/2026-03-09T1842Z/logs/train_phase7_grounded_tokenonly_canary_detectfix_20260306_100356.nohup with huggingface_hub
Browse files
pre_redeploy/2026-03-09T1842Z/logs/train_phase7_grounded_tokenonly_canary_detectfix_20260306_100356.nohup
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[THREADS] CPU cores=16, torch.get_num_threads()=16
|
| 2 |
+
2026-03-06 10:03:57.267 | DEBUG | ttnn:<module>:77 - Initial ttnn.CONFIG:
|
| 3 |
+
Config{cache_path=/root/.cache/ttnn,model_cache_path=/root/.cache/ttnn/models,tmp_dir=/tmp/ttnn,enable_model_cache=false,enable_fast_runtime_mode=true,throw_exception_on_fallback=false,enable_logging=false,enable_graph_report=false,enable_detailed_buffer_report=false,enable_detailed_tensor_report=false,enable_comparison_mode=false,comparison_mode_should_raise_exception=false,comparison_mode_pcc=0.9999,root_report_path=generated/ttnn/reports,report_name=std::nullopt,std::nullopt}
|
| 4 |
+
[TT-OPS] FORCE_TT_LINEAR_BWD=1 (CPU fallback disabled for TTLinear backward)
|
| 5 |
+
[TT-OPS] EXPERIMENTAL_TT_SDPA_BWD=1 (manual SDPA backward enabled)
|
| 6 |
+
[DATA] Found 2 JSONL files in /root/newtie/training_data/phase7
|
| 7 |
+
[POWER] phase7_grounded_tokenonly_canary_detectfix: eval_interval=301
|
| 8 |
+
[HARDWARE] 2× N300S detected (4 Wormhole chips)
|
| 9 |
+
96 GB VRAM across 4 chips (12 GB each)
|
| 10 |
+
All matmul compute → Wormhole Tensix cores
|
| 11 |
+
CPU only holds parameters for optimizer — zero CPU compute
|
| 12 |
+
[TOKENIZER] Byte-level fallback (no sentencepiece)
|
| 13 |
+
[DATA] Loaded /root/newtie/training_data/phase7/sft_p7_grounded_agency_generated_20260306.jsonl: 408 total entries
|
| 14 |
+
[DATA] Loaded /root/newtie/training_data/phase7/sft_p7_grounded_agency_seed_20260306.jsonl: 420 total entries
|
| 15 |
+
[DATA] Total: 420 training entries
|
| 16 |
+
[TT-MESH] adaptive plan tokens=8192 ops=262 2-chip=70 4-chip=0 8-chip=192 est_speedup_vs_static8=1.005x
|
| 17 |
+
[TT-MESH] layers.0.norm1 -> 2 (small_control_op)
|
| 18 |
+
[TT-MESH] layers.0.attn.q_proj -> 8 (large_kernel)
|
| 19 |
+
[TT-MESH] layers.0.attn.out_proj -> 8 (large_kernel)
|
| 20 |
+
[TT-MESH] layers.0.ffn.gate_up -> 8 (throughput_projection)
|
| 21 |
+
[TT-MESH] layers.4.moe.router -> 2 (small_control_op)
|
| 22 |
+
[TT-MESH] layers.4.moe.expert_up -> 8 (moe_routing_hot_path)
|
| 23 |
+
|
| 24 |
+
[MODEL] Initializing Newtie V5...
|
| 25 |
+
[SUPERBYTE] SuperByte-Lite enabled (patch=8, stride=4, pool=avg)
|
| 26 |
+
Total params: 355,431,103 (355.4M)
|
| 27 |
+
Active/token: 213,873,343 (213.9M, 60.2%)
|
| 28 |
+
BF16 size: 0.71 GB
|
| 29 |
+
[DEVICE] Model params on CPU (for AdamW), compute → Wormhole chips
|
| 30 |
+
[MEMORY] WARNING: Gradient checkpointing DISABLED — may OOM!
|
| 31 |
+
[HARDWARE] pipeline profile: cards=2, chips=4, stages=4, layers_per_stage=8
|
| 32 |
+
2026-03-06 10:04:03.210 | info | UMD | Established firmware bundle version: 19.4.2 (topology_discovery.cpp:418)
|