Add files using upload-large-folder tool

LTA_openwebtext_dualt/logs/infer_owt_compact_v2048_latest236k_anchor_state_steps128_n8_large_20260520_211237.log

@@ -0,0 +1,199 @@
+[infer] gpu=0 label=c256_low cmax=256 temps=1.00,1.05,1.10
+[infer] gpu=1 label=c256_high cmax=256 temps=1.12,1.15,1.20,1.30,1.45
+[infer] gpu=3 label=c1024_high cmax=1024 temps=1.12,1.15,1.20,1.30,1.45
+[infer] gpu=2 label=c1024_low cmax=1024 temps=1.00,1.05,1.10
+[ckpt] runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt step=238000
+[ckpt] runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt step=238000
+[decode-base] n=8 max_len=1024 steps=128 model_t=flow
+[ckpt] runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt step=238000
+[decode-base] n=8 max_len=1024 steps=128 model_t=flow
+[decode-time] schedule=linear s=[0.0,0.25] gumbel=(2.2,0.8) force_final=True t0=0.000000 t_mid=0.500000 t_end=1.000000 dt_mean=0.007812 dt_max=0.007812
+[decode-time] schedule=linear s=[0.0,0.25] gumbel=(2.2,0.8) force_final=True t0=0.000000 t_mid=0.500000 t_end=1.000000 dt_mean=0.007812 dt_max=0.007812
+[ckpt] runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt step=238000
+[decode-base] n=8 max_len=1024 steps=128 model_t=flow
+[decode-base] n=8 max_len=1024 steps=128 model_t=flow
+[decode-time] schedule=linear s=[0.0,0.25] gumbel=(2.2,0.8) force_final=True t0=0.000000 t_mid=0.500000 t_end=1.000000 dt_mean=0.007812 dt_max=0.007812
+[decode-time] schedule=linear s=[0.0,0.25] gumbel=(2.2,0.8) force_final=True t0=0.000000 t_mid=0.500000 t_end=1.000000 dt_mean=0.007812 dt_max=0.007812
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.12 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.00 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.15 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.05 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.20 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.10 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.0, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 2.5961969121729997, "nll_per_token": 0.9540476481119792, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 2.6498743397101228, "nll_per_token": 0.9745122198965035, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.9879212243443718, "unique_tokens": 69, "token_count": 8192, "distinct_1": 0.0084228515625, "distinct_2": 0.026637341153470186, "top_token_mass": 0.1795654296875}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.0, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 3.592485981997712, "nll_per_token": 1.2788444369446998, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 3.929751877309959, "nll_per_token": 1.3685762883434978, "tokens": 1983, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 2.143849744436887, "unique_tokens": 105, "token_count": 8192, "distinct_1": 0.0128173828125, "distinct_2": 0.0448435972629521, "top_token_mass": 0.1473388671875}}
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.05, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 3.0506829418331822, "nll_per_token": 1.1153654809091604, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 3.441284508071593, "nll_per_token": 1.2358448051967503, "tokens": 1956, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.831910997598835, "unique_tokens": 61, "token_count": 8192, "distinct_1": 0.0074462890625, "distinct_2": 0.022116324535679376, "top_token_mass": 0.2073974609375}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.05, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 5.223055260936073, "nll_per_token": 1.6530825297037761, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 6.6517307134400525, "nll_per_token": 1.8948770784667712, "tokens": 1903, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 2.4435827826077756, "unique_tokens": 209, "token_count": 8192, "distinct_1": 0.0255126953125, "distinct_2": 0.07795698924731183, "top_token_mass": 0.2362060546875}}
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.1, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 2.6042242951698436, "nll_per_token": 0.9571348554947797, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 2.5352120024490743, "nll_per_token": 0.9302772637037655, "tokens": 1956, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.7218896892009548, "unique_tokens": 87, "token_count": 8192, "distinct_1": 0.0106201171875, "distinct_2": 0.036779081133919846, "top_token_mass": 0.2508544921875}}
+[done] docs/lta_samples/metrics_20260520/owt_compact_v2048_latest236k_anchor_state_steps128_n8_large/c1024_low.jsonl
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.1, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 3.3091695321797516, "nll_per_token": 1.1966972612867168, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 3.352661334301717, "nll_per_token": 1.2097544585957247, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.6081862093608303, "unique_tokens": 163, "token_count": 8192, "distinct_1": 0.0198974609375, "distinct_2": 0.05755131964809384, "top_token_mass": 0.1328125}}
+[done] docs/lta_samples/metrics_20260520/owt_compact_v2048_latest236k_anchor_state_steps128_n8_large/c256_low.jsonl
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.30 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 1/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 2/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 3/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 4/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 5/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 6/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 7/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[decode] temp=1.45 final=state rule=dual_line_resample support=1 semantic=1 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise time_path=0.0000->1.0000 generated 8/8
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.12, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 2.8236783740208375, "nll_per_token": 1.038040422925762, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 3.075275397771778, "nll_per_token": 1.123394457499186, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.7479719031636645, "unique_tokens": 88, "token_count": 8192, "distinct_1": 0.0107421875, "distinct_2": 0.03250244379276637, "top_token_mass": 0.1514892578125}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.12, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 3.066636236743537, "nll_per_token": 1.120581272536633, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 3.2168138105157573, "nll_per_token": 1.1683913698383406, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.589027264097043, "unique_tokens": 121, "token_count": 8192, "distinct_1": 0.0147705078125, "distinct_2": 0.05535190615835777, "top_token_mass": 0.25}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.15, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 2.765000473413227, "nll_per_token": 1.0170408061906404, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 2.5985446365924196, "nll_per_token": 0.9549515331492705, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.3535927406560724, "unique_tokens": 74, "token_count": 8192, "distinct_1": 0.009033203125, "distinct_2": 0.02541544477028348, "top_token_mass": 0.143310546875}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.15, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 3.8504377975579187, "nll_per_token": 1.348186855486193, "tokens": 1907, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 3.6229466339708805, "nll_per_token": 1.2872876820144696, "tokens": 1807, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.0184736955969358, "unique_tokens": 48, "token_count": 8192, "distinct_1": 0.005859375, "distinct_2": 0.02126099706744868, "top_token_mass": 0.2606201171875}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.2, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 1.9771231385360284, "nll_per_token": 0.681642827800676, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 1.742214032336782, "nll_per_token": 0.5551567367478913, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.2931185887482526, "unique_tokens": 56, "token_count": 8192, "distinct_1": 0.0068359375, "distinct_2": 0.020527859237536656, "top_token_mass": 0.142822265625}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.2, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 2.789709523279108, "nll_per_token": 1.0259374768930525, "tokens": 1953, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 2.680677056263793, "nll_per_token": 0.9860693955506903, "tokens": 1953, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 0.7400165432226921, "unique_tokens": 31, "token_count": 8192, "distinct_1": 0.0037841796875, "distinct_2": 0.012952101661779082, "top_token_mass": 0.4510498046875}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.3, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 4.189420015894294, "nll_per_token": 1.432562303325357, "tokens": 1752, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 4.068326236907239, "nll_per_token": 1.4032316708673624, "tokens": 1752, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 0.7696244616605461, "unique_tokens": 26, "token_count": 8192, "distinct_1": 0.003173828125, "distinct_2": 0.018206256109481917, "top_token_mass": 0.3785400390625}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.3, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 8.604361010102021, "nll_per_token": 2.152269168928558, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 8.285594906578865, "nll_per_token": 2.1145184535606236, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 1.3249730869579166, "unique_tokens": 96, "token_count": 8192, "distinct_1": 0.01171875, "distinct_2": 0.06048387096774194, "top_token_mass": 0.4315185546875}}
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.45, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 5.979892476558285, "nll_per_token": 1.788402587292241, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 5.979892476558285, "nll_per_token": 1.788402587292241, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 0.7126024592246277, "unique_tokens": 21, "token_count": 8192, "distinct_1": 0.0025634765625, "distinct_2": 0.009164222873900294, "top_token_mass": 0.465576171875}}
+[done] docs/lta_samples/metrics_20260520/owt_compact_v2048_latest236k_anchor_state_steps128_n8_large/c1024_high.jsonl
+[summary] {"type": "summary", "checkpoint": "runs/lta_owt_compact_gpt2bpe_v2048_stream1024_fullycoupled_rmsnorm_nobias_adamw_wd0p1_logitnormal_m1p5_s0p8_hardce_mask1p0-1p0_fp32_ddit768x12_gbs512_8gpu_1m_20260517_141027/latest.pt", "step": 238000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_time_schedule": "linear", "decode_s_min_frac": 0.0, "decode_s_max_frac": 0.25, "decode_force_final_t": true, "decode_time_grid": [0.0, 0.0078125, 0.015625, 0.0234375, 0.03125, 0.0390625, 0.046875, 0.0546875, 0.0625, 0.0703125, 0.078125, 0.0859375, 0.09375, 0.1015625, 0.109375, 0.1171875, 0.125, 0.1328125, 0.140625, 0.1484375, 0.15625, 0.1640625, 0.171875, 0.1796875, 0.1875, 0.1953125, 0.203125, 0.2109375, 0.21875, 0.2265625, 0.234375, 0.2421875, 0.25, 0.2578125, 0.265625, 0.2734375, 0.28125, 0.2890625, 0.296875, 0.3046875, 0.3125, 0.3203125, 0.328125, 0.3359375, 0.34375, 0.3515625, 0.359375, 0.3671875, 0.375, 0.3828125, 0.390625, 0.3984375, 0.40625, 0.4140625, 0.421875, 0.4296875, 0.4375, 0.4453125, 0.453125, 0.4609375, 0.46875, 0.4765625, 0.484375, 0.4921875, 0.5, 0.5078125, 0.515625, 0.5234375, 0.53125, 0.5390625, 0.546875, 0.5546875, 0.5625, 0.5703125, 0.578125, 0.5859375, 0.59375, 0.6015625, 0.609375, 0.6171875, 0.625, 0.6328125, 0.640625, 0.6484375, 0.65625, 0.6640625, 0.671875, 0.6796875, 0.6875, 0.6953125, 0.703125, 0.7109375, 0.71875, 0.7265625, 0.734375, 0.7421875, 0.75, 0.7578125, 0.765625, 0.7734375, 0.78125, 0.7890625, 0.796875, 0.8046875, 0.8125, 0.8203125, 0.828125, 0.8359375, 0.84375, 0.8515625, 0.859375, 0.8671875, 0.875, 0.8828125, 0.890625, 0.8984375, 0.90625, 0.9140625, 0.921875, 0.9296875, 0.9375, 0.9453125, 0.953125, 0.9609375, 0.96875, 0.9765625, 0.984375, 0.9921875, 1.0], "decode_rule": "dual_line_resample", "support_power": 1.0, "semantic_power": 1.0, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 256.0, "target_prob": 1.0, "endpoint_temp": 1.45, "final_from": "state", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 8, "seed": 20260520}, "raw_genppl": {"ppl": 58.72244210463748, "nll_per_token": 4.072821972416897, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 56.63366130772546, "nll_per_token": 4.036603531182981, "tokens": 2040, "kept_samples": 8, "total_samples": 8, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 2.1996299837879643, "unique_tokens": 152, "token_count": 8192, "distinct_1": 0.0185546875, "distinct_2": 0.14100684261974586, "top_token_mass": 0.2628173828125}}
+[done] docs/lta_samples/metrics_20260520/owt_compact_v2048_latest236k_anchor_state_steps128_n8_large/c256_high.jsonl
+docs/lta_samples/metrics_20260520/owt_compact_v2048_latest236k_anchor_state_steps128_n8_large/compact_metrics.csv
+cmax	temp	raw	strip	H	unique	top	tokens
+256	1.00	3.5925	3.9298	2.1438	105	0.1473	2040
+256	1.05	5.2231	6.6517	2.4436	209	0.2362	2040
+256	1.10	3.3092	3.3527	1.6082	163	0.1328	2040
+256	1.12	3.0666	3.2168	1.5890	121	0.2500	2040
+256	1.15	3.8504	3.6229	1.0185	48	0.2606	1907
+256	1.20	2.7897	2.6807	0.7400	31	0.4510	1953
+256	1.30	8.6044	8.2856	1.3250	96	0.4315	2040
+256	1.45	58.7224	56.6337	2.1996	152	0.2628	2040
+1024	1.00	2.5962	2.6499	1.9879	69	0.1796	2040
+1024	1.05	3.0507	3.4413	1.8319	61	0.2074	2040
+1024	1.10	2.6042	2.5352	1.7219	87	0.2509	2040
+1024	1.12	2.8237	3.0753	1.7480	88	0.1515	2040
+1024	1.15	2.7650	2.5985	1.3536	74	0.1433	2040
+1024	1.20	1.9771	1.7422	1.2931	56	0.1428	2040
+1024	1.30	4.1894	4.0683	0.7696	26	0.3785	1752
+1024	1.45	5.9799	5.9799	0.7126	21	0.4656	2040
+
+[best raw<50 sorted entropy]
+c256 temp=1.05	raw=5.2231	strip=6.6517	H=2.4436	unique=209	top=0.2362
+c256 temp=1.00	raw=3.5925	strip=3.9298	H=2.1438	unique=105	top=0.1473
+c1024 temp=1.00	raw=2.5962	strip=2.6499	H=1.9879	unique=69	top=0.1796
+c1024 temp=1.05	raw=3.0507	strip=3.4413	H=1.8319	unique=61	top=0.2074
+c1024 temp=1.12	raw=2.8237	strip=3.0753	H=1.7480	unique=88	top=0.1515
+c1024 temp=1.10	raw=2.6042	strip=2.5352	H=1.7219	unique=87	top=0.2509
+c256 temp=1.10	raw=3.3092	strip=3.3527	H=1.6082	unique=163	top=0.1328
+c256 temp=1.12	raw=3.0666	strip=3.2168	H=1.5890	unique=121	top=0.2500
+c1024 temp=1.15	raw=2.7650	strip=2.5985	H=1.3536	unique=74	top=0.1433
+c256 temp=1.30	raw=8.6044	strip=8.2856	H=1.3250	unique=96	top=0.4315
+c1024 temp=1.20	raw=1.9771	strip=1.7422	H=1.2931	unique=56	top=0.1428
+c256 temp=1.15	raw=3.8504	strip=3.6229	H=1.0185	unique=48	top=0.2606
+c1024 temp=1.30	raw=4.1894	strip=4.0683	H=0.7696	unique=26	top=0.3785
+c256 temp=1.20	raw=2.7897	strip=2.6807	H=0.7400	unique=31	top=0.4510
+c1024 temp=1.45	raw=5.9799	strip=5.9799	H=0.7126	unique=21	top=0.4656

LTA_openwebtext_dualt/logs/owt_classic_fullvocab_len1024_16gpu/lta_owt_classic_fullvocab_bert_c1024_len1024_lr2e4_gbs2048_2node8gpu_1m_save10k_t-20260522043432-f7vrv.node1.log

@@ -0,0 +1,701 @@
+W0521 20:35:15.179000 10381 torch/distributed/run.py:792] 
+W0521 20:35:15.179000 10381 torch/distributed/run.py:792] *****************************************
+W0521 20:35:15.179000 10381 torch/distributed/run.py:792] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
+W0521 20:35:15.179000 10381 torch/distributed/run.py:792] *****************************************
+[W521 20:35:15.706532924 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.907715488 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.157873989 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.201304866 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.219422947 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.221040686 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.221680401 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.229310067 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+[W521 20:35:20.236203327 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
+t-20260522043432-f7vrv-worker-1:10470:10470 [0] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10470:10470 [0] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10470:10470 [0] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10470:10470 [0] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10470:10470 [0] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10471:10471 [1] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10471:10471 [1] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10471:10471 [1] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10471:10471 [1] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10471:10471 [1] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10473:10473 [3] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10473:10473 [3] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10473:10473 [3] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10473:10473 [3] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10473:10473 [3] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10474:10474 [4] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10474:10474 [4] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10474:10474 [4] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10474:10474 [4] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10474:10474 [4] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10475:10475 [5] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10475:10475 [5] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10475:10475 [5] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10475:10475 [5] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10475:10475 [5] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10476:10476 [6] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10476:10476 [6] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10476:10476 [6] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10476:10476 [6] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10472:10472 [2] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10472:10472 [2] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10472:10472 [2] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10472:10472 [2] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10476:10476 [6] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10472:10472 [2] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10477:10477 [7] NCCL INFO cudaDriverVersion 12080
+t-20260522043432-f7vrv-worker-1:10477:10477 [7] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10477:10477 [7] NCCL INFO Bootstrap: Using eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10477:10477 [7] NCCL INFO NCCL version 2.25.1+cuda12.8
+t-20260522043432-f7vrv-worker-1:10477:10477 [7] NCCL INFO Comm config Blocking set to 1
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO ncclCommInitRankConfig comm 0x8764a40 rank 8 nranks 16 cudaDev 0 nvmlDev 0 busId 65040 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO P2P plugin v9 IBext_v9
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.48<0>
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO Using network IBext_v9
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO ncclCommInitRankConfig comm 0x9d44240 rank 9 nranks 16 cudaDev 1 nvmlDev 1 busId 67020 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO ncclCommInitRankConfig comm 0xafe2000 rank 12 nranks 16 cudaDev 4 nvmlDev 4 busId 6f020 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO ncclCommInitRankConfig comm 0xb79f100 rank 11 nranks 16 cudaDev 3 nvmlDev 3 busId 6b020 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO ncclCommInitRankConfig comm 0xb2b9cc0 rank 13 nranks 16 cudaDev 5 nvmlDev 5 busId 71020 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO ncclCommInitRankConfig comm 0xa052ac0 rank 14 nranks 16 cudaDev 6 nvmlDev 6 busId 73020 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO ncclCommInitRankConfig comm 0x9c2d640 rank 10 nranks 16 cudaDev 2 nvmlDev 2 busId 69020 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO ncclCommInitRankConfig comm 0xa1eb800 rank 15 nranks 16 cudaDev 7 nvmlDev 7 busId 75020 commId 0xb1911eddb57862e3 - Init START
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO RAS client listening socket at ::1<28028>
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO Bootstrap timings total 0.103273 (create 0.000023, send 0.000196, recv 0.000464, ring 0.102079, delay 0.000001)
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO Bootstrap timings total 0.107654 (create 0.000020, send 0.000244, recv 0.004616, ring 0.102490, delay 0.000001)
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO Bootstrap timings total 0.119285 (create 0.000020, send 0.000206, recv 0.011956, ring 0.106797, delay 0.000000)
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO Bootstrap timings total 0.106209 (create 0.000020, send 0.000266, recv 0.000499, ring 0.105137, delay 0.000001)
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO Bootstrap timings total 0.136282 (create 0.000033, send 0.000274, recv 0.017298, ring 0.118082, delay 0.000001)
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO Bootstrap timings total 0.131916 (create 0.000022, send 0.000246, recv 0.000411, ring 0.105106, delay 0.000001)
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO Bootstrap timings total 0.355903 (create 0.000023, send 0.000319, recv 0.249960, ring 0.105247, delay 0.000001)
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO Bootstrap timings total 1.028546 (create 0.000027, send 0.000495, recv 0.673228, ring 0.006598, delay 0.000001)
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO MNNVL busId 0x75020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO MNNVL busId 0x73020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO MNNVL busId 0x69020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO MNNVL busId 0x71020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO MNNVL busId 0x6f020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO MNNVL busId 0x6b020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO MNNVL busId 0x67020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO MNNVL busId 0x65040 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO Setting affinity for GPU 1 to 03ffffff,ffffffff,ffffffff
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO NVLS multicast support is available on dev 1
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO Setting affinity for GPU 7 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO Setting affinity for GPU 5 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO NVLS multicast support is available on dev 5
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO NVLS multicast support is available on dev 7
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO Setting affinity for GPU 3 to 03ffffff,ffffffff,ffffffff
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO Setting affinity for GPU 0 to 03ffffff,ffffffff,ffffffff
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO NVLS multicast support is available on dev 3
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO Setting affinity for GPU 2 to 03ffffff,ffffffff,ffffffff
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO NVLS multicast support is available on dev 0
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO Setting affinity for GPU 6 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO NVLS multicast support is available on dev 6
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO Setting affinity for GPU 4 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO NVLS multicast support is available on dev 4
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO NVLS multicast support is available on dev 2
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO comm 0xa1eb800 rank 15 nRanks 16 nNodes 2 localRanks 8 localRank 7 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO comm 0xa052ac0 rank 14 nRanks 16 nNodes 2 localRanks 8 localRank 6 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO comm 0xb2b9cc0 rank 13 nRanks 16 nNodes 2 localRanks 8 localRank 5 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO comm 0xafe2000 rank 12 nRanks 16 nNodes 2 localRanks 8 localRank 4 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO comm 0xb79f100 rank 11 nRanks 16 nNodes 2 localRanks 8 localRank 3 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO comm 0x9c2d640 rank 10 nRanks 16 nNodes 2 localRanks 8 localRank 2 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO comm 0x9d44240 rank 9 nRanks 16 nNodes 2 localRanks 8 localRank 1 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO comm 0x8764a40 rank 8 nRanks 16 nNodes 2 localRanks 8 localRank 0 MNNVL 0
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO Trees [0] -1/-1/-1->15->14 [1] 8/-1/-1->15->14 [2] 8/-1/-1->15->14 [3] 8/-1/-1->15->14 [4] 8/-1/-1->15->14 [5] 8/-1/-1->15->14 [6] 8/-1/-1->15->14 [7] 8/-1/-1->15->7 [8] -1/-1/-1->15->14 [9] 8/-1/-1->15->14 [10] 8/-1/-1->15->14 [11] 8/-1/-1->15->14 [12] 8/-1/-1->15->14 [13] 8/-1/-1->15->14 [14] 8/-1/-1->15->14 [15] 8/7/-1->15->-1
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO Trees [0] 14/-1/-1->13->12 [1] 14/-1/-1->13->12 [2] 14/-1/-1->13->12 [3] 14/-1/-1->13->12 [4] 14/-1/-1->13->12 [5] 14/-1/-1->13->5 [6] -1/-1/-1->13->12 [7] 14/-1/-1->13->12 [8] 14/-1/-1->13->12 [9] 14/-1/-1->13->12 [10] 14/-1/-1->13->12 [11] 14/-1/-1->13->12 [12] 14/-1/-1->13->12 [13] 14/5/-1->13->-1 [14] -1/-1/-1->13->12 [15] 14/-1/-1->13->12
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO Trees [0] 15/-1/-1->14->13 [1] 15/-1/-1->14->13 [2] 15/-1/-1->14->13 [3] 15/-1/-1->14->13 [4] 15/-1/-1->14->13 [5] 15/-1/-1->14->13 [6] 15/-1/-1->14->6 [7] -1/-1/-1->14->13 [8] 15/-1/-1->14->13 [9] 15/-1/-1->14->13 [10] 15/-1/-1->14->13 [11] 15/-1/-1->14->13 [12] 15/-1/-1->14->13 [13] 15/-1/-1->14->13 [14] 15/6/-1->14->-1 [15] -1/-1/-1->14->13
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO Trees [0] 13/-1/-1->12->11 [1] 13/-1/-1->12->11 [2] 13/-1/-1->12->11 [3] 13/-1/-1->12->11 [4] 13/-1/-1->12->4 [5] -1/-1/-1->12->11 [6] 13/-1/-1->12->11 [7] 13/-1/-1->12->11 [8] 13/-1/-1->12->11 [9] 13/-1/-1->12->11 [10] 13/-1/-1->12->11 [11] 13/-1/-1->12->11 [12] 13/4/-1->12->-1 [13] -1/-1/-1->12->11 [14] 13/-1/-1->12->11 [15] 13/-1/-1->12->11
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO Trees [0] 10/-1/-1->9->8 [1] 10/-1/-1->9->1 [2] -1/-1/-1->9->8 [3] 10/-1/-1->9->8 [4] 10/-1/-1->9->8 [5] 10/-1/-1->9->8 [6] 10/-1/-1->9->8 [7] 10/-1/-1->9->8 [8] 10/-1/-1->9->8 [9] 10/1/-1->9->-1 [10] -1/-1/-1->9->8 [11] 10/-1/-1->9->8 [12] 10/-1/-1->9->8 [13] 10/-1/-1->9->8 [14] 10/-1/-1->9->8 [15] 10/-1/-1->9->8
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO Trees [0] 11/-1/-1->10->9 [1] 11/-1/-1->10->9 [2] 11/-1/-1->10->2 [3] -1/-1/-1->10->9 [4] 11/-1/-1->10->9 [5] 11/-1/-1->10->9 [6] 11/-1/-1->10->9 [7] 11/-1/-1->10->9 [8] 11/-1/-1->10->9 [9] 11/-1/-1->10->9 [10] 11/2/-1->10->-1 [11] -1/-1/-1->10->9 [12] 11/-1/-1->10->9 [13] 11/-1/-1->10->9 [14] 11/-1/-1->10->9 [15] 11/-1/-1->10->9
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO Trees [0] 12/-1/-1->11->10 [1] 12/-1/-1->11->10 [2] 12/-1/-1->11->10 [3] 12/-1/-1->11->3 [4] -1/-1/-1->11->10 [5] 12/-1/-1->11->10 [6] 12/-1/-1->11->10 [7] 12/-1/-1->11->10 [8] 12/-1/-1->11->10 [9] 12/-1/-1->11->10 [10] 12/-1/-1->11->10 [11] 12/3/-1->11->-1 [12] -1/-1/-1->11->10 [13] 12/-1/-1->11->10 [14] 12/-1/-1->11->10 [15] 12/-1/-1->11->10
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO Trees [0] 9/-1/-1->8->0 [1] -1/-1/-1->8->15 [2] 9/-1/-1->8->15 [3] 9/-1/-1->8->15 [4] 9/-1/-1->8->15 [5] 9/-1/-1->8->15 [6] 9/-1/-1->8->15 [7] 9/-1/-1->8->15 [8] 9/0/-1->8->-1 [9] -1/-1/-1->8->15 [10] 9/-1/-1->8->15 [11] 9/-1/-1->8->15 [12] 9/-1/-1->8->15 [13] 9/-1/-1->8->15 [14] 9/-1/-1->8->15 [15] 9/-1/-1->8->15
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO P2P Chunksize set to 131072
+t-20260522043432-f7vrv-worker-1:10471:10623 [1] NCCL INFO [Proxy Service] Device 1 CPU core 10
+t-20260522043432-f7vrv-worker-1:10471:10626 [1] NCCL INFO [Proxy Service UDS] Device 1 CPU core 12
+t-20260522043432-f7vrv-worker-1:10475:10622 [5] NCCL INFO [Proxy Service] Device 5 CPU core 173
+t-20260522043432-f7vrv-worker-1:10475:10627 [5] NCCL INFO [Proxy Service UDS] Device 5 CPU core 175
+t-20260522043432-f7vrv-worker-1:10470:10625 [0] NCCL INFO [Proxy Service] Device 0 CPU core 60
+t-20260522043432-f7vrv-worker-1:10473:10628 [3] NCCL INFO [Proxy Service] Device 3 CPU core 2
+t-20260522043432-f7vrv-worker-1:10474:10630 [4] NCCL INFO [Proxy Service] Device 4 CPU core 108
+t-20260522043432-f7vrv-worker-1:10472:10629 [2] NCCL INFO [Proxy Service UDS] Device 2 CPU core 82
+t-20260522043432-f7vrv-worker-1:10470:10632 [0] NCCL INFO [Proxy Service UDS] Device 0 CPU core 62
+t-20260522043432-f7vrv-worker-1:10472:10624 [2] NCCL INFO [Proxy Service] Device 2 CPU core 80
+t-20260522043432-f7vrv-worker-1:10477:10631 [7] NCCL INFO [Proxy Service] Device 7 CPU core 92
+t-20260522043432-f7vrv-worker-1:10473:10633 [3] NCCL INFO [Proxy Service UDS] Device 3 CPU core 4
+t-20260522043432-f7vrv-worker-1:10474:10635 [4] NCCL INFO [Proxy Service UDS] Device 4 CPU core 110
+t-20260522043432-f7vrv-worker-1:10476:10634 [6] NCCL INFO [Proxy Service] Device 6 CPU core 94
+t-20260522043432-f7vrv-worker-1:10477:10636 [7] NCCL INFO [Proxy Service UDS] Device 7 CPU core 96
+t-20260522043432-f7vrv-worker-1:10476:10637 [6] NCCL INFO [Proxy Service UDS] Device 6 CPU core 98
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO ncclCommInitRankConfig comm 0x8764a40 rank 8 nranks 16 cudaDev 0 nvmlDev 0 busId 65040 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO ncclCommInitRankConfig comm 0xafe2000 rank 12 nranks 16 cudaDev 4 nvmlDev 4 busId 6f020 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO ncclCommInitRankConfig comm 0xa052ac0 rank 14 nranks 16 cudaDev 6 nvmlDev 6 busId 73020 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO ncclCommInitRankConfig comm 0x9d44240 rank 9 nranks 16 cudaDev 1 nvmlDev 1 busId 67020 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO ncclCommInitRankConfig comm 0xb2b9cc0 rank 13 nranks 16 cudaDev 5 nvmlDev 5 busId 71020 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO ncclCommInitRankConfig comm 0x9c2d640 rank 10 nranks 16 cudaDev 2 nvmlDev 2 busId 69020 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
+t-20260522043432-f7vrv-worker-1:10470:10519 [0] NCCL INFO Init timings - ncclCommInitRankConfig: rank 8 nranks 16 total 2.74 (kernels 0.20, alloc 0.23, bootstrap 1.03, allgathers 0.01, topo 0.56, graphs 0.01, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO ncclCommInitRankConfig comm 0xb79f100 rank 11 nranks 16 cudaDev 3 nvmlDev 3 busId 6b020 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10474:10530 [4] NCCL INFO Init timings - ncclCommInitRankConfig: rank 12 nranks 16 total 2.49 (kernels 0.22, alloc 0.85, bootstrap 0.14, allgathers 0.00, topo 0.56, graphs 0.02, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10476:10542 [6] NCCL INFO Init timings - ncclCommInitRankConfig: rank 14 nranks 16 total 2.45 (kernels 0.20, alloc 0.85, bootstrap 0.11, allgathers 0.00, topo 0.56, graphs 0.02, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO ncclCommInitRankConfig comm 0xa1eb800 rank 15 nranks 16 cudaDev 7 nvmlDev 7 busId 75020 commId 0xb1911eddb57862e3 - Init COMPLETE
+t-20260522043432-f7vrv-worker-1:10471:10520 [1] NCCL INFO Init timings - ncclCommInitRankConfig: rank 9 nranks 16 total 2.61 (kernels 0.30, alloc 0.67, bootstrap 0.36, allgathers 0.01, topo 0.56, graphs 0.01, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10475:10532 [5] NCCL INFO Init timings - ncclCommInitRankConfig: rank 13 nranks 16 total 2.47 (kernels 0.21, alloc 0.85, bootstrap 0.12, allgathers 0.01, topo 0.56, graphs 0.01, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10472:10543 [2] NCCL INFO Init timings - ncclCommInitRankConfig: rank 10 nranks 16 total 2.44 (kernels 0.20, alloc 0.85, bootstrap 0.11, allgathers 0.01, topo 0.56, graphs 0.01, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10473:10522 [3] NCCL INFO Init timings - ncclCommInitRankConfig: rank 11 nranks 16 total 2.56 (kernels 0.29, alloc 0.86, bootstrap 0.13, allgathers 0.01, topo 0.56, graphs 0.01, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10477:10544 [7] NCCL INFO Init timings - ncclCommInitRankConfig: rank 15 nranks 16 total 2.43 (kernels 0.21, alloc 0.83, bootstrap 0.10, allgathers 0.00, topo 0.56, graphs 0.02, connections 0.71, rest 0.00)
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 02/0 : 8[0] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 04/0 : 8[0] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 06/0 : 8[0] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 10/0 : 8[0] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 12/0 : 8[0] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 14/0 : 8[0] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 00/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 00/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 02/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 02/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 00/0 : 12[4] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 00/0 : 14[6] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 04/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 04/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 02/0 : 12[4] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 00/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 00/0 : 10[2] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 02/0 : 14[6] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 06/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 06/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 06/0 : 12[4] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 02/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 04/0 : 10[2] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 04/0 : 14[6] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 08/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 08/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 08/0 : 12[4] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 04/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 06/0 : 10[2] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 08/0 : 14[6] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 10/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 10/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 10/0 : 12[4] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 06/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 08/0 : 10[2] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 10/0 : 14[6] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 12/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 12/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 14/0 : 12[4] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 08/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 12/0 : 10[2] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 12/0 : 14[6] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 14/0 : 13[5] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 14/0 : 9[1] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 10/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 14/0 : 10[2] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 01/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 12/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 03/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 14/0 : 11[3] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 05/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 07/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 09/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 11/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 13/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 15/0 : 8[0] -> 15[7] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10646 [1] NCCL INFO [Proxy Progress] Device 1 CPU core 12
+t-20260522043432-f7vrv-worker-1:10476:10647 [6] NCCL INFO [Proxy Progress] Device 6 CPU core 96
+t-20260522043432-f7vrv-worker-1:10475:10648 [5] NCCL INFO [Proxy Progress] Device 5 CPU core 175
+t-20260522043432-f7vrv-worker-1:10474:10649 [4] NCCL INFO [Proxy Progress] Device 4 CPU core 108
+t-20260522043432-f7vrv-worker-1:10473:10650 [3] NCCL INFO [Proxy Progress] Device 3 CPU core 4
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 01/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 09/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 00/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Channel 08/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10651 [7] NCCL INFO [Proxy Progress] Device 7 CPU core 94
+t-20260522043432-f7vrv-worker-1:10470:10652 [0] NCCL INFO [Proxy Progress] Device 0 CPU core 62
+t-20260522043432-f7vrv-worker-1:10472:10653 [2] NCCL INFO [Proxy Progress] Device 2 CPU core 82
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 00/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 08/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 01/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 09/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 03/0 : 9[1] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 05/0 : 9[1] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 07/0 : 9[1] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 11/0 : 9[1] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 13/0 : 9[1] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Channel 15/0 : 9[1] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 04/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 05/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 02/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 06/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 03/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 07/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 10/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 12/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 14/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 15/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 11/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 13/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 04/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 06/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 05/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 07/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 02/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 03/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 15/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 13/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 12/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 14/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 10/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 11/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 00/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 02/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 04/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 06/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 08/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 10/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 01/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 12/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 01/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 01/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 03/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 14/0 : 15[7] -> 8[0] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 03/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 03/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 05/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 01/0 : 11[3] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 01/0 : 13[5] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 01/0 : 15[7] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 05/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 05/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 07/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 05/0 : 11[3] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 03/0 : 13[5] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 03/0 : 15[7] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 07/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 07/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 09/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 07/0 : 11[3] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 07/0 : 13[5] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 05/0 : 15[7] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 09/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 09/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 11/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 09/0 : 11[3] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 09/0 : 13[5] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 09/0 : 15[7] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 11/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 11/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 13/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 13/0 : 11[3] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 11/0 : 13[5] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 11/0 : 15[7] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 13/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 13/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Channel 15/0 : 12[4] -> 11[3] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Channel 15/0 : 11[3] -> 10[2] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Channel 15/0 : 13[5] -> 12[4] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Channel 13/0 : 15[7] -> 14[6] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Channel 15/0 : 14[6] -> 13[5] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Channel 15/0 : 10[2] -> 9[1] via P2P/CUMEM
+t-20260522043432-f7vrv-worker-1:10472:10624 [2] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10476:10634 [6] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10472:10624 [2] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10472:10624 [2] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10476:10634 [6] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10476:10634 [6] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10471:10623 [1] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10471:10623 [1] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10471:10623 [1] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10477:10631 [7] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10477:10631 [7] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10477:10631 [7] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10474:10630 [4] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10474:10630 [4] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10474:10630 [4] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10470:10625 [0] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10470:10625 [0] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10470:10625 [0] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10475:10622 [5] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10475:10622 [5] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10475:10622 [5] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10473:10628 [3] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10473:10628 [3] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
+t-20260522043432-f7vrv-worker-1:10473:10628 [3] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
+t-20260522043432-f7vrv-worker-1:10477:10645 [7] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10476:10641 [6] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10475:10642 [5] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10474:10638 [4] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10473:10644 [3] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10472:10640 [2] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10471:10643 [1] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10470:10639 [0] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO NVLS comm 0x9c2d640 headRank 2 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO NVLS comm 0xa052ac0 headRank 6 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO NVLS comm 0xafe2000 headRank 4 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO NVLS comm 0x8764a40 headRank 0 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO NVLS comm 0x9d44240 headRank 1 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO NVLS comm 0xb79f100 headRank 3 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO NVLS comm 0xa1eb800 headRank 7 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO NVLS comm 0xb2b9cc0 headRank 5 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 01/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 00/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 00/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 00/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 00/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 00/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 00/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 00/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 02/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 01/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 01/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 01/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 02/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 01/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 01/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 03/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 03/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 03/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 02/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 01/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 02/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 02/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 02/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 04/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 03/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 04/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 04/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 03/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 03/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 03/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 02/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 05/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 04/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 05/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 04/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 05/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 05/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 04/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 04/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 06/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 06/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 05/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 06/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 06/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 05/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 06/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 07/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 05/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 06/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 07/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 07/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 07/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 07/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 07/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 08/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 08/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 09/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 08/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 08/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 08/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 06/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 08/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 09/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 10/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 10/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 09/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 09/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 09/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 07/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 09/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 10/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 11/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 11/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 10/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 10/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 10/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 08/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 11/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 12/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 12/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 11/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 11/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 11/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 11/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 09/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 12/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 13/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 13/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 12/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 12/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 12/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 13/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 14/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 14/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 14/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 13/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 13/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 13/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 10/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 15/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 14/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 15/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 15/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 15/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 14/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 14/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 00/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 12/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 01/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 15/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 00/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 00/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 00/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 01/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 15/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 02/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 01/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 00/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 02/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 13/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 01/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 02/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 02/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 03/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 00/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 03/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 02/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 01/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 14/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 03/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 04/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 03/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 03/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 04/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 01/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 02/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 05/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 15/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 04/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 05/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 04/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 05/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 02/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 03/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 06/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 00/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 06/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 05/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 05/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 06/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 04/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 04/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 07/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 06/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 07/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 07/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 07/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 01/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 08/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 05/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 06/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 08/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 08/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 08/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 09/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 03/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 09/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 06/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 07/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 09/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 10/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 09/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 10/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 10/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 04/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 07/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 08/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 10/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 11/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 10/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 11/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 11/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 05/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 08/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 09/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 11/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 12/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 11/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 12/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 13/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 09/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 10/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 06/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 12/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 12/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 13/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 13/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 14/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 10/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 13/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 07/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 11/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 13/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 14/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 14/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Channel 15/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Channel 14/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 12/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Channel 15/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Channel 15/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Channel 15/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 12/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 08/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 14/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 13/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 09/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Channel 15/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 14/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 11/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Channel 15/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 12/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 13/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 14/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Channel 15/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
+t-20260522043432-f7vrv-worker-1:10475:10764 [5] NCCL INFO Connected NVLS tree
+t-20260522043432-f7vrv-worker-1:10477:10763 [7] NCCL INFO Connected NVLS tree
+t-20260522043432-f7vrv-worker-1:10474:10759 [4] NCCL INFO Connected NVLS tree
+t-20260522043432-f7vrv-worker-1:10476:10758 [6] NCCL INFO Connected NVLS tree
+t-20260522043432-f7vrv-worker-1:10471:10762 [1] NCCL INFO Connected NVLS tree
+t-20260522043432-f7vrv-worker-1:10472:10757 [2] NCCL INFO Connected NVLS tree
+t-20260522043432-f7vrv-worker-1:10470:10760 [0] NCCL INFO Connected NVLS tree
+t-20260522043432-f7vrv-worker-1:10473:10761 [3] NCCL INFO Connected NVLS tree

LTA_openwebtext_dualt/logs/owt_fully_uncertain_schedule_step115k_n64.log

@@ -0,0 +1,62 @@
+[forbid_endpoint_ids] n=352 first=[94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125]
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top16_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top16_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 16, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top16_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top16_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 16, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top16_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top16_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 16, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top32_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top32_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 32, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top32_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top32_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 32, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top32_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow0p75_noise0_state_sample_uncertain_ft0p9_top32_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 32, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 63.76480697133721, "sample_entropy": 4.33102141624885, "distinct_1": 0.0581512451171875, "distinct_2": 0.4148796432062561, "top_token_mass": 0.1015472412109375, "tokens_scored": 58094, "readability_score": 4.25096603664177, "mean_chars": 3898.96875, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top16_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top16_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top16_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top16_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top16_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top16_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top32_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top32_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top32_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top32_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top32_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p7_top32_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top16_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top16_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 16, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top16_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top16_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 16, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top16_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top16_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 16, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top32_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top32_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 32, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top32_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top32_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 32, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top32_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_linear_tpow1p25_noise0_state_sample_uncertain_ft0p9_top32_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "linear", "t_power": 1.25, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.9, "final_top_k": 32, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 66.85183152738117, "sample_entropy": 4.356138008348162, "distinct_1": 0.05511474609375, "distinct_2": 0.3989338954056696, "top_token_mass": 0.11016845703125, "tokens_scored": 57555, "readability_score": 4.42523594915668, "mean_chars": 3833.359375, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "cosine", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 64.07011369477127, "sample_entropy": 4.356511462923227, "distinct_1": 0.0601043701171875, "distinct_2": 0.42505193059628543, "top_token_mass": 0.0958251953125, "tokens_scored": 58493, "readability_score": 4.207463999796301, "mean_chars": 3923.03125, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "cosine", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 64.07011369477127, "sample_entropy": 4.356511462923227, "distinct_1": 0.0601043701171875, "distinct_2": 0.42505193059628543, "top_token_mass": 0.0958251953125, "tokens_scored": 58493, "readability_score": 4.207463999796301, "mean_chars": 3923.03125, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top16_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "cosine", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 16, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 64.07011369477127, "sample_entropy": 4.356511462923227, "distinct_1": 0.0601043701171875, "distinct_2": 0.42505193059628543, "top_token_mass": 0.0958251953125, "tokens_scored": 58493, "readability_score": 4.207463999796301, "mean_chars": 3923.03125, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p55_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p55_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "cosine", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.55, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 64.07011369477127, "sample_entropy": 4.356511462923227, "distinct_1": 0.0601043701171875, "distinct_2": 0.42505193059628543, "top_token_mass": 0.0958251953125, "tokens_scored": 58493, "readability_score": 4.207463999796301, "mean_chars": 3923.03125, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p7_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p7_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "cosine", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.7, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 64.07011369477127, "sample_entropy": 4.356511462923227, "distinct_1": 0.0601043701171875, "distinct_2": 0.42505193059628543, "top_token_mass": 0.0958251953125, "tokens_scored": 58493, "readability_score": 4.207463999796301, "mean_chars": 3923.03125, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p85_anchored
+[summary] {"name": "steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p7_top32_th0p85_anchored", "step": 116000, "n_samples": 64, "steps": 48, "concentration_max": 256.0, "temp_start": 1.3, "temp_end": 1.15, "temp_schedule": "cosine", "t_power": 0.75, "eta0": 0.0, "eta_schedule": "none", "noise_conc": 1.0, "final_from": "state", "final_decode": "sample_uncertain", "final_temp": 0.7, "final_top_k": 32, "final_uncertain_threshold": 0.85, "update_rule": "anchored", "model_t_mode": "post", "lock_bos": true, "lock_final_eos": false, "detok_genppl": 64.07011369477127, "sample_entropy": 4.356511462923227, "distinct_1": 0.0601043701171875, "distinct_2": 0.42505193059628543, "top_token_mass": 0.0958251953125, "tokens_scored": 58493, "readability_score": 4.207463999796301, "mean_chars": 3923.03125, "replacement_chars": 0.0}
+[decode] steps48_c256_mtpost_t1p3to1p15_cosine_tpow0p75_noise0_state_sample_uncertain_ft0p9_top16_th0p55_anchored

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/accesstype.f90

@@ -0,0 +1,13 @@
+module foo
+  public
+  type, private, bind(c) :: a
+     integer :: i
+  end type a
+  type, bind(c) :: b_
+     integer :: j
+  end type b_
+  public :: b_
+  type :: c
+     integer :: k
+  end type c
+end module foo

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/data_multiplier.f

@@ -0,0 +1,5 @@
+      BLOCK DATA MYBLK
+      IMPLICIT DOUBLE PRECISION (A-H,O-Z)
+      COMMON /MYCOM/ IVAR1, IVAR2, IVAR3, IVAR4, EVAR5
+            DATA IVAR1, IVAR2, IVAR3, IVAR4, EVAR5 /2*3,2*2,0.0D0/
+      END

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/data_with_comments.f

@@ -0,0 +1,8 @@
+      BLOCK DATA PARAM_INI
+      COMMON /MYCOM/ MYTAB
+      INTEGER  MYTAB(3)
+      DATA MYTAB/
+     *   0, ! 1 and more commenty stuff
+     *   4, ! 2
+     *   0 /
+      END

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/foo_deps.f90

@@ -0,0 +1,6 @@
+module foo
+  type bar
+    character(len = 4) :: text
+  end type bar
+  type(bar), parameter :: abar = bar('abar')
+end module foo

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/gh22648.pyf

@@ -0,0 +1,7 @@
+python module iri16py ! in
+    interface  ! in :iri16py
+        block data  ! in :iri16py:iridreg_modified.for
+           COMMON /fircom/ eden,tabhe,tabla,tabmo,tabza,tabfl
+       end block data 
+    end interface 
+end python module iri16py

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/gh23879.f90

@@ -0,0 +1,20 @@
+module gh23879
+    implicit none
+    private
+    public :: foo
+
+ contains
+
+    subroutine foo(a, b)
+       integer, intent(in) :: a
+       integer, intent(out) :: b
+       b = a
+       call bar(b)
+    end subroutine
+
+    subroutine bar(x)
+        integer, intent(inout) :: x
+        x = 2*x
+     end subroutine
+
+ end module gh23879

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/operators.f90

@@ -0,0 +1,49 @@
+module foo
+  type bar
+     character(len = 32) :: item
+  end type bar
+  interface operator(.item.)
+     module procedure item_int, item_real
+  end interface operator(.item.)
+  interface operator(==)
+     module procedure items_are_equal
+  end interface operator(==)
+  interface assignment(=)
+     module procedure get_int, get_real
+  end interface assignment(=)
+contains
+  function item_int(val) result(elem)
+    integer, intent(in) :: val
+    type(bar) :: elem
+
+    write(elem%item, "(I32)") val
+  end function item_int
+
+  function item_real(val) result(elem)
+    real, intent(in) :: val
+    type(bar) :: elem
+
+    write(elem%item, "(1PE32.12)") val
+  end function item_real
+
+  function items_are_equal(val1, val2) result(equal)
+    type(bar), intent(in) :: val1, val2
+    logical :: equal
+
+    equal = (val1%item == val2%item)
+  end function items_are_equal
+
+  subroutine get_real(rval, item)
+    real, intent(out) :: rval
+    type(bar), intent(in) :: item
+
+    read(item%item, *) rval
+  end subroutine get_real
+
+  subroutine get_int(rval, item)
+    integer, intent(out) :: rval
+    type(bar), intent(in) :: item
+
+    read(item%item, *) rval
+  end subroutine get_int
+end module foo

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/privatemod.f90

@@ -0,0 +1,11 @@
+module foo
+  private
+  integer :: a
+  public :: setA
+  integer :: b
+contains
+  subroutine setA(v)
+    integer, intent(in) :: v
+    a = v
+  end subroutine setA
+end module foo

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/f2py/tests/src/crackfortran/pubprivmod.f90

@@ -0,0 +1,10 @@
+module foo
+  public
+  integer, private :: a
+  integer :: b
+contains
+  subroutine setA(v)
+    integer, intent(in) :: v
+    a = v
+  end subroutine setA
+end module foo

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/dinov3_vit/configuration_dinov3_vit.py

@@ -0,0 +1,111 @@
+# Copyright 2025 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""DINOv3 model configuration"""
+
+from huggingface_hub.dataclasses import strict
+
+from ...backbone_utils import BackboneConfigMixin
+from ...configuration_utils import PreTrainedConfig
+from ...utils import auto_docstring
+
+
+@auto_docstring(checkpoint="facebook/dinov3-vits16-pretrain-lvd1689m")
+@strict
+class DINOv3ViTConfig(BackboneConfigMixin, PreTrainedConfig):
+    r"""
+    rope_theta (`float`, *optional*, defaults to 100.0):
+        The base period of the RoPE embeddings.
+    query_bias (`bool`, *optional*, defaults to `True`):
+        Whether to add a bias to the query projection.
+    key_bias (`bool`, *optional*, defaults to `False`):
+        Whether to add a bias to the key projection.
+    value_bias (`bool`, *optional*, defaults to `True`):
+        Whether to add a bias to the value projection.
+    proj_bias (`bool`, *optional*, defaults to `True`):
+        Whether to add a bias to the output projection.
+    layerscale_value (`float`, *optional*, defaults to 1.0):
+        Initial value to use for layer scale.
+    use_gated_mlp (`bool`, *optional*, defaults to `False`):
+        Whether to use the SwiGLU feedforward neural network.
+    num_register_tokens (`int`, *optional*, defaults to 0):
+        The number of register tokens.
+    pos_embed_shift (`float`, *optional*):
+        Amount to randomly shift position embedding coordinates in [-shift, shift],
+        applied only in training mode if not `None`.
+    pos_embed_jitter (`float`, *optional*):
+        Amount to randomly jitter position embedding coordinates in log-uniform value in [1/jitter, jitter],
+        applied only in training mode if not `None`.
+    pos_embed_rescale (`float`, *optional*, defaults to 2.0):
+        Amount to randomly rescale position embedding coordinates in log-uniform value in [1/rescale, rescale],
+        applied only in training mode if not `None`.
+    apply_layernorm (`bool`, *optional*, defaults to `True`):
+        Whether to apply layer normalization to the feature maps when used as backbone.
+    reshape_hidden_states (`bool`, *optional*, defaults to `True`):
+        Whether to reshape the hidden states to spatial dimensions when used as backbone.
+
+    Example:
+
+    ```python
+    >>> from transformers import DINOv3ViTConfig, DINOv3ViTModel
+
+    >>> # Initializing a DINOv3 ViT-small style configuration
+    >>> config = DINOv3ViTConfig()
+
+    >>> # Initializing a model (with random weights) from the config
+    >>> model = DINOv3ViTModel(config)
+
+    >>> # Accessing the model config
+    >>> config = model.config
+    ```"""
+
+    model_type = "dinov3_vit"
+
+    patch_size: int | list[int] | tuple[int, int] = 16
+    hidden_size: int = 384
+    intermediate_size: int = 1536
+    num_hidden_layers: int = 12
+    num_attention_heads: int = 6
+    hidden_act: str = "gelu"
+    attention_dropout: float | int = 0.0
+    initializer_range: float = 0.02
+    layer_norm_eps: float = 1e-5
+    rope_theta: float = 100.0
+    image_size: int | list[int] | tuple[int, int] = 224
+    num_channels: int = 3
+    query_bias: bool = True
+    key_bias: bool = False
+    value_bias: bool = True
+    proj_bias: bool = True
+    mlp_bias: bool = True
+    layerscale_value: float = 1.0
+    drop_path_rate: float | int = 0.0
+    use_gated_mlp: bool = False
+    num_register_tokens: int = 0
+    pos_embed_shift: float | None = None
+    pos_embed_jitter: float | None = None
+    pos_embed_rescale: float | None = 2.0
+    _out_features: list[str] | None = None
+    _out_indices: list[int] | None = None
+    apply_layernorm: bool = True
+    reshape_hidden_states: bool = True
+
+    def __post_init__(self, **kwargs):
+        self.stage_names = ["stem"] + [f"stage{i}" for i in range(1, self.num_hidden_layers + 1)]
+        self.set_output_features_output_indices(
+            out_indices=kwargs.pop("out_indices", None), out_features=kwargs.pop("out_features", None)
+        )
+        super().__post_init__(**kwargs)
+
+
+__all__ = ["DINOv3ViTConfig"]

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/dinov3_vit/image_processing_dinov3_vit.py

@@ -0,0 +1,89 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Image processor class for DINOv3."""
+
+import torch
+from torchvision.transforms.v2 import functional as tvF
+
+from ...image_processing_backends import TorchvisionBackend
+from ...image_processing_utils import BatchFeature
+from ...image_transforms import group_images_by_shape, reorder_images
+from ...image_utils import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, PILImageResampling, SizeDict
+from ...utils import (
+    TensorType,
+    auto_docstring,
+    logging,
+)
+
+
+logger = logging.get_logger(__name__)
+
+
+@auto_docstring
+class DINOv3ViTImageProcessor(TorchvisionBackend):
+    resample = PILImageResampling.BILINEAR
+    image_mean = IMAGENET_DEFAULT_MEAN
+    image_std = IMAGENET_DEFAULT_STD
+    size = {"height": 224, "width": 224}
+    do_resize = True
+    do_rescale = True
+    do_normalize = True
+
+    # Overridden for DINOv3 to preserve order of transforms
+    # rescale -> resize -> normalize
+    def _preprocess(
+        self,
+        images: list["torch.Tensor"],
+        do_resize: bool,
+        size: SizeDict,
+        resample: "PILImageResampling | tvF.InterpolationMode | int | None",
+        do_center_crop: bool,
+        crop_size: SizeDict,
+        do_rescale: bool,
+        rescale_factor: float,
+        do_normalize: bool,
+        image_mean: float | list[float] | None,
+        image_std: float | list[float] | None,
+        disable_grouping: bool | None,
+        return_tensors: str | TensorType | None,
+        **kwargs,
+    ) -> BatchFeature:
+        # Group images by size for batched resizing
+        grouped_images, grouped_images_index = group_images_by_shape(images, disable_grouping=disable_grouping)
+        resized_images_grouped = {}
+        for shape, stacked_images in grouped_images.items():
+            if do_rescale:
+                stacked_images = self.rescale(stacked_images, rescale_factor)
+            if do_resize:
+                stacked_images = self.resize(image=stacked_images, size=size, resample=resample, antialias=True)
+            resized_images_grouped[shape] = stacked_images
+        resized_images = reorder_images(resized_images_grouped, grouped_images_index)
+
+        # Group images by size for further processing
+        # Needed in case do_resize is False, or resize returns images with different sizes
+        grouped_images, grouped_images_index = group_images_by_shape(resized_images, disable_grouping=disable_grouping)
+        processed_images_grouped = {}
+        for shape, stacked_images in grouped_images.items():
+            if do_center_crop:
+                stacked_images = self.center_crop(stacked_images, crop_size)
+            if do_normalize:
+                stacked_images = self.normalize(stacked_images, image_mean, image_std)
+            processed_images_grouped[shape] = stacked_images
+
+        processed_images = reorder_images(processed_images_grouped, grouped_images_index)
+
+        return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
+
+
+__all__ = ["DINOv3ViTImageProcessor"]

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/dinov3_vit/modeling_dinov3_vit.py

@@ -0,0 +1,630 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/dinov3_vit/modular_dinov3_vit.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_dinov3_vit.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# Copyright 2025 Meta AI and The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+from collections.abc import Callable
+from dataclasses import dataclass
+
+import numpy as np
+import torch
+from torch import nn
+
+from ... import initialization as init
+from ...activations import ACT2FN
+from ...backbone_utils import BackboneMixin, filter_output_hidden_states
+from ...modeling_layers import GradientCheckpointingLayer
+from ...modeling_outputs import BackboneOutput, BaseModelOutput, BaseModelOutputWithPooling
+from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from ...processing_utils import Unpack
+from ...pytorch_utils import compile_compatible_method_lru_cache
+from ...utils import TransformersKwargs, auto_docstring
+from ...utils.generic import can_return_tuple, maybe_autocast, merge_with_config_defaults
+from ...utils.output_capturing import capture_outputs
+from .configuration_dinov3_vit import DINOv3ViTConfig
+
+
+@auto_docstring(
+    custom_intro="""
+    Output type of [`DINOv3ViTBackbone`], extending [`BackboneOutput`] with optional CLS tokens from
+    each selected feature stage (used when `config.return_class_token=True`).
+    """
+)
+@dataclass
+class DINOv3ViTBackboneOutput(BackboneOutput):
+    r"""
+    cls_tokens (`tuple(torch.FloatTensor)`, *optional*):
+        CLS token from each selected feature stage, each of shape `(batch_size, hidden_size)`.
+        Only present when `config.return_class_token=True`.
+    """
+
+    cls_tokens: tuple[torch.FloatTensor] | None = None
+
+
+class DINOv3ViTEmbeddings(nn.Module):
+    """
+    Construct the CLS token, mask token, position and patch embeddings.
+    """
+
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__()
+        self.config = config
+        self.cls_token = nn.Parameter(torch.randn(1, 1, config.hidden_size))
+        self.mask_token = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
+        self.register_tokens = nn.Parameter(torch.empty(1, config.num_register_tokens, config.hidden_size))
+        self.patch_embeddings = nn.Conv2d(
+            config.num_channels, config.hidden_size, kernel_size=config.patch_size, stride=config.patch_size
+        )
+
+    def forward(self, pixel_values: torch.Tensor, bool_masked_pos: torch.Tensor | None = None) -> torch.Tensor:
+        batch_size = pixel_values.shape[0]
+        target_dtype = self.patch_embeddings.weight.dtype
+
+        # (batch_size, num_channels, height, width) -> (batch_size, num_patches, hidden_size)
+        patch_embeddings = self.patch_embeddings(pixel_values.to(dtype=target_dtype))
+        patch_embeddings = patch_embeddings.flatten(2).transpose(1, 2)
+
+        if bool_masked_pos is not None:
+            mask_token = self.mask_token.to(patch_embeddings.dtype)
+            patch_embeddings = torch.where(bool_masked_pos.unsqueeze(-1), mask_token, patch_embeddings)
+
+        # Add CLS and register tokens
+        cls_token = self.cls_token.expand(batch_size, -1, -1)
+        register_tokens = self.register_tokens.expand(batch_size, -1, -1)
+        embeddings = torch.cat([cls_token, register_tokens, patch_embeddings], dim=1)
+
+        return embeddings
+
+
+@compile_compatible_method_lru_cache(maxsize=32)
+def get_patches_center_coordinates(
+    num_patches_h: int, num_patches_w: int, dtype: torch.dtype, device: torch.device
+) -> torch.Tensor:
+    """
+    Computes the 2D coordinates of the centers of image patches, normalized to the range [-1, +1].
+    The center of each patch is exactly halfway between its top-left and bottom-right corners.
+
+    Args:
+        num_patches_h (int): Number of patches along the vertical (height) axis.
+        num_patches_w (int): Number of patches along the horizontal (width) axis.
+        dtype (torch.dtype): The desired data type of the returned tensor.
+
+    Returns:
+        torch.Tensor: A tensor of shape (height * width, 2), where each row contains the (y, x)
+            coordinates of a patch center, normalized to [-1, +1].
+    """
+    coords_h = torch.arange(0.5, num_patches_h, dtype=dtype, device=device)
+    coords_w = torch.arange(0.5, num_patches_w, dtype=dtype, device=device)
+    coords_h = coords_h / num_patches_h
+    coords_w = coords_w / num_patches_w
+    # (height, width, 2) -> (height * width, 2)
+    coords = torch.stack(torch.meshgrid(coords_h, coords_w, indexing="ij"), dim=-1)
+    coords = coords.flatten(0, 1)
+    # Shift range [0, 1] to [-1, +1]
+    coords = 2.0 * coords - 1.0
+    return coords
+
+
+def augment_patches_center_coordinates(
+    coords: torch.Tensor,
+    shift: float | None = None,
+    jitter: float | None = None,
+    rescale: float | None = None,
+) -> torch.Tensor:
+    # Shift coords by adding a uniform value in [-shift, shift]
+    if shift is not None:
+        shift_hw = torch.empty((1, 2), device=coords.device, dtype=coords.dtype)
+        shift_hw = shift_hw.uniform_(-shift, shift)
+        coords = coords + shift_hw
+
+    # Jitter coords by multiplying the range [-1, 1] by a log-uniform value in [1/jitter, jitter]
+    if jitter is not None:
+        jitter_range = np.log(jitter)
+        jitter_hw = torch.empty((1, 2), device=coords.device, dtype=coords.dtype)
+        jitter_hw = jitter_hw.uniform_(-jitter_range, jitter_range).exp()
+        coords = coords * jitter_hw
+
+    # Rescale coords by multiplying the range [-1, 1] by a log-uniform value in [1/rescale, rescale]
+    if rescale is not None:
+        rescale_range = np.log(rescale)
+        rescale_hw = torch.empty(1, device=coords.device, dtype=coords.dtype)
+        rescale_hw = rescale_hw.uniform_(-rescale_range, rescale_range).exp()
+        coords = coords * rescale_hw
+
+    return coords
+
+
+class DINOv3ViTRopePositionEmbedding(nn.Module):
+    inv_freq: torch.Tensor
+
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__()
+
+        self.config = config
+        self.base = config.rope_theta
+        self.head_dim = config.hidden_size // config.num_attention_heads
+        self.num_patches_h = config.image_size // config.patch_size
+        self.num_patches_w = config.image_size // config.patch_size
+
+        inv_freq = 1 / self.base ** torch.arange(0, 1, 4 / self.head_dim, dtype=torch.float32)  # (head_dim / 4,)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+    def forward(self, pixel_values: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
+        _, _, height, width = pixel_values.shape
+        num_patches_h = height // self.config.patch_size
+        num_patches_w = width // self.config.patch_size
+
+        device = pixel_values.device
+        device_type = device.type if isinstance(device.type, str) and device.type != "mps" else "cpu"
+
+        with maybe_autocast(device_type=device_type, enabled=False):  # Force float32
+            # Although we could precompute static patch_coords from image_size and patch_size in the config,
+            # the model was trained with random_scale, so it can process images of varying sizes.
+            # Therefore, it's better to compute patch_coords dynamically (with lru_cache).
+            patch_coords = get_patches_center_coordinates(
+                num_patches_h, num_patches_w, dtype=torch.float32, device=device
+            )
+            if self.training:
+                patch_coords = augment_patches_center_coordinates(
+                    patch_coords,
+                    shift=self.config.pos_embed_shift,
+                    jitter=self.config.pos_embed_jitter,
+                    rescale=self.config.pos_embed_rescale,
+                )
+
+            # (height * width, 2, head_dim / 4) -> (height * width, head_dim / 2) -> (height * width, head_dim)
+            angles = 2 * math.pi * patch_coords[:, :, None] * self.inv_freq[None, None, :]
+            angles = angles.flatten(1, 2)
+            angles = angles.tile(2)
+
+            cos = torch.cos(angles)
+            sin = torch.sin(angles)
+
+        dtype = pixel_values.dtype
+        return cos.to(dtype=dtype), sin.to(dtype=dtype)
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: torch.Tensor | None,
+    scaling: float | None = None,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    if scaling is None:
+        scaling = query.size(-1) ** -0.5
+
+    # Take the dot product between "query" and "key" to get the raw attention scores.
+    attn_weights = torch.matmul(query, key.transpose(2, 3)) * scaling
+
+    if attention_mask is not None:
+        attn_weights = attn_weights + attention_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+
+    attn_output = torch.matmul(attn_weights, value)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+def apply_rotary_pos_emb(
+    q: torch.Tensor, k: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor, **kwargs
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """Applies Rotary Position Embedding to the query and key tensors, but only to the patch tokens,
+    ignoring the prefix tokens (cls token and register tokens).
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+
+    num_tokens = q.shape[-2]
+    num_patches = sin.shape[-2]
+    num_prefix_tokens = num_tokens - num_patches  # cls token + register tokens
+
+    q_prefix_tokens, q_patches = q.split((num_prefix_tokens, num_patches), dim=-2)
+    k_prefix_tokens, k_patches = k.split((num_prefix_tokens, num_patches), dim=-2)
+
+    # apply rope only to patch tokens
+    q_patches = (q_patches * cos) + (rotate_half(q_patches) * sin)
+    k_patches = (k_patches * cos) + (rotate_half(k_patches) * sin)
+
+    q = torch.cat((q_prefix_tokens, q_patches), dim=-2)
+    k = torch.cat((k_prefix_tokens, k_patches), dim=-2)
+
+    return q, k
+
+
+class DINOv3ViTAttention(nn.Module):
+    """
+    Multi-headed attention compatible with ALL_ATTENTION_FUNCTIONS.
+    """
+
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__()
+        self.config = config
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_heads
+        self.is_causal = False
+
+        self.scaling = self.head_dim**-0.5
+        self.is_causal = False
+
+        self.dropout = config.attention_dropout
+        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.key_bias)
+        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.value_bias)
+
+        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.query_bias)
+        self.o_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.proj_bias)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple[torch.Tensor, torch.Tensor | None]:
+        """Input shape: Batch x Time x Channel"""
+
+        batch_size, patches, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(batch_size, patches, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, attn_weights
+
+
+class DINOv3ViTLayerScale(nn.Module):
+    def __init__(self, config) -> None:
+        super().__init__()
+        self.lambda1 = nn.Parameter(config.layerscale_value * torch.ones(config.hidden_size))
+
+    def forward(self, hidden_state: torch.Tensor) -> torch.Tensor:
+        return hidden_state * self.lambda1
+
+
+class DINOv3ViTMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.mlp_bias)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.up_proj(x)))
+
+
+class DINOv3ViTGatedMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.mlp_bias)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+
+
+class Dinov3ViTDropPath(nn.Module):
+    """Stochastic depth (DropPath) per sample, for residual blocks.
+
+    Identity when ``drop_prob`` is 0 or outside training. See `Deep Networks with Stochastic Depth
+    <https://arxiv.org/abs/1603.09382>`_.
+    """
+
+    def __init__(self, drop_prob: float = 0.0) -> None:
+        super().__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        if self.drop_prob == 0.0 or not self.training:
+            return hidden_states
+        keep_prob = 1 - self.drop_prob
+        shape = (hidden_states.shape[0],) + (1,) * (hidden_states.ndim - 1)
+        random_tensor = torch.rand(shape, dtype=hidden_states.dtype, device=hidden_states.device)
+        random_tensor = torch.floor(random_tensor + keep_prob)
+        return hidden_states.div(keep_prob) * random_tensor
+
+    def extra_repr(self) -> str:
+        return f"p={self.drop_prob}"
+
+
+class DINOv3ViTLayer(GradientCheckpointingLayer):
+    """This corresponds to the Block class in the original implementation."""
+
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__()
+
+        self.norm1 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.attention = DINOv3ViTAttention(config)
+        self.layer_scale1 = DINOv3ViTLayerScale(config)
+        self.drop_path = Dinov3ViTDropPath(config.drop_path_rate) if config.drop_path_rate > 0.0 else nn.Identity()
+
+        self.norm2 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+        if config.use_gated_mlp:
+            self.mlp = DINOv3ViTGatedMLP(config)
+        else:
+            self.mlp = DINOv3ViTMLP(config)
+        self.layer_scale2 = DINOv3ViTLayerScale(config)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.Tensor:
+        # Attention with residual connection
+        residual = hidden_states
+        hidden_states = self.norm1(hidden_states)
+        hidden_states, _ = self.attention(
+            hidden_states,
+            attention_mask=attention_mask,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = self.layer_scale1(hidden_states)
+        hidden_states = self.drop_path(hidden_states) + residual
+
+        # MLP with residual connection
+        residual = hidden_states
+        hidden_states = self.norm2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = self.layer_scale2(hidden_states)
+        hidden_states = self.drop_path(hidden_states) + residual
+
+        return hidden_states
+
+
+@auto_docstring
+class DINOv3ViTPreTrainedModel(PreTrainedModel):
+    config: DINOv3ViTConfig
+    base_model_prefix = "model"
+    main_input_name = "pixel_values"
+    input_modalities = ("image",)
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["DINOv3ViTLayer"]
+    _supports_sdpa = True
+    _supports_flash_attn = True
+    _supports_flex_attn = True
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "hidden_states": DINOv3ViTLayer,
+        "attentions": DINOv3ViTAttention,
+    }
+
+    @torch.no_grad()
+    def _init_weights(self, module) -> None:
+        """Initialize the weights"""
+        if isinstance(module, (nn.Linear, nn.Conv2d)):
+            init.trunc_normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                init.zeros_(module.bias)
+        elif isinstance(module, nn.LayerNorm):
+            init.zeros_(module.bias)
+            init.ones_(module.weight)
+        elif isinstance(module, DINOv3ViTEmbeddings):
+            init.trunc_normal_(module.cls_token, mean=0.0, std=self.config.initializer_range)
+            if module.config.num_register_tokens > 0:
+                init.trunc_normal_(module.register_tokens, mean=0.0, std=self.config.initializer_range)
+            init.zeros_(module.mask_token)
+        elif isinstance(module, DINOv3ViTLayerScale):
+            init.constant_(module.lambda1, self.config.layerscale_value)
+        elif isinstance(module, DINOv3ViTRopePositionEmbedding):
+            inv_freq = 1 / module.base ** torch.arange(0, 1, 4 / module.head_dim, dtype=torch.float32)
+            init.copy_(module.inv_freq, inv_freq)
+
+
+class DINOv3ViTEncoder(DINOv3ViTPreTrainedModel):
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__(config)
+        self.layer = nn.ModuleList([DINOv3ViTLayer(config) for _ in range(config.num_hidden_layers)])
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @merge_with_config_defaults
+    @capture_outputs(tie_last_hidden_states=False)
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutput:
+        for layer_module in self.layer:
+            hidden_states = layer_module(hidden_states, position_embeddings=position_embeddings, **kwargs)
+
+        return BaseModelOutput(last_hidden_state=hidden_states)
+
+
+@auto_docstring
+class DINOv3ViTModel(DINOv3ViTPreTrainedModel):
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__(config)
+        self.embeddings = DINOv3ViTEmbeddings(config)
+        self.rope_embeddings = DINOv3ViTRopePositionEmbedding(config)
+        self.model = DINOv3ViTEncoder(config)
+        self.norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings.patch_embeddings
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        pixel_values: torch.Tensor,
+        bool_masked_pos: torch.Tensor | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutputWithPooling:
+        r"""
+        bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, sequence_length)`):
+            Boolean masked positions. Indicates which patches are masked (1) and which aren't (0). Only relevant for
+            pre-training.
+        """
+
+        pixel_values = pixel_values.to(self.embeddings.patch_embeddings.weight.dtype)
+        hidden_states = self.embeddings(pixel_values, bool_masked_pos=bool_masked_pos)
+        position_embeddings = self.rope_embeddings(pixel_values)
+
+        output = self.model(hidden_states, position_embeddings, **kwargs)
+        sequence_output = self.norm(output.last_hidden_state)
+        pooled_output = sequence_output[:, 0, :]
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            hidden_states=output.hidden_states,
+            attentions=output.attentions,
+        )
+
+
+@auto_docstring
+class DINOv3ViTBackbone(BackboneMixin, DINOv3ViTPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embeddings = DINOv3ViTEmbeddings(config)
+        self.rope_embeddings = DINOv3ViTRopePositionEmbedding(config)
+        self.model = DINOv3ViTEncoder(config)
+        self.norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.gradient_checkpointing = False
+
+        self.num_features = [config.hidden_size for _ in range(config.num_hidden_layers + 1)]
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings.patch_embeddings
+
+    @can_return_tuple
+    @filter_output_hidden_states
+    @auto_docstring
+    def forward(
+        self,
+        pixel_values: torch.Tensor,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> DINOv3ViTBackboneOutput:
+        pixel_values = pixel_values.to(self.embeddings.patch_embeddings.weight.dtype)
+        hidden_states = self.embeddings(pixel_values)
+        position_embeddings = self.rope_embeddings(pixel_values)
+
+        kwargs["output_hidden_states"] = True  # required to extract layers for the stages
+        output = self.model(hidden_states, position_embeddings, **kwargs)
+        stage_hidden_states = output.hidden_states
+
+        batch_size, _, image_height, image_width = pixel_values.shape
+        patch_size = self.config.patch_size
+        num_patches_height = image_height // patch_size
+        num_patches_width = image_width // patch_size
+
+        num_prefix = 1 + getattr(self.config, "num_register_tokens", 0)
+        return_class_token = getattr(self.config, "return_class_token", False)
+
+        feature_maps, cls_tokens = [], []
+        sequence_output = None
+        last_stage_idx = len(self.stage_names) - 1
+        for idx, (stage_name, hidden_state) in enumerate(zip(self.stage_names, stage_hidden_states)):
+            if idx == last_stage_idx:
+                hidden_state = self.norm(hidden_state)
+                sequence_output = hidden_state
+            elif self.config.apply_layernorm:
+                hidden_state = self.norm(hidden_state)
+
+            if stage_name in self.out_features:
+                if return_class_token:
+                    cls_tokens.append(hidden_state[:, 0, :])
+                patch_tokens = hidden_state[:, num_prefix:, :]
+                if self.config.reshape_hidden_states:
+                    fmap = (
+                        patch_tokens.reshape(batch_size, num_patches_height, num_patches_width, patch_tokens.shape[-1])
+                        .permute(0, 3, 1, 2)
+                        .contiguous()
+                    )
+                else:
+                    fmap = patch_tokens
+
+                feature_maps.append(fmap)
+
+        output = DINOv3ViTBackboneOutput(
+            feature_maps=tuple(feature_maps),
+            cls_tokens=tuple(cls_tokens) if return_class_token else None,
+            hidden_states=output.hidden_states,
+            attentions=output.attentions,
+        )
+        output.last_hidden_state = sequence_output
+
+        return output
+
+
+__all__ = ["DINOv3ViTModel", "DINOv3ViTPreTrainedModel", "DINOv3ViTBackbone"]

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/dinov3_vit/modular_dinov3_vit.py

@@ -0,0 +1,532 @@
+# Copyright 2025 Meta AI and The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""PyTorch DINOv3 model."""
+
+import math
+from collections.abc import Callable
+from dataclasses import dataclass
+
+import numpy as np
+import torch
+from torch import nn
+
+from transformers.models.arcee.modeling_arcee import ArceeMLP
+from transformers.models.dinov2.modeling_dinov2 import (
+    Dinov2LayerScale,
+    Dinov2PreTrainedModel,
+    eager_attention_forward,
+)
+from transformers.models.llama.modeling_llama import LlamaMLP
+from transformers.models.pixtral.modeling_pixtral import PixtralAttention, rotate_half
+
+from ... import initialization as init
+from ...backbone_utils import BackboneMixin, filter_output_hidden_states
+from ...modeling_layers import GradientCheckpointingLayer
+from ...modeling_outputs import BackboneOutput, BaseModelOutput, BaseModelOutputWithPooling
+from ...modeling_utils import ALL_ATTENTION_FUNCTIONS
+from ...processing_utils import Unpack
+from ...pytorch_utils import compile_compatible_method_lru_cache
+from ...utils import TransformersKwargs, auto_docstring, logging
+from ...utils.generic import can_return_tuple, maybe_autocast, merge_with_config_defaults
+from ...utils.output_capturing import capture_outputs
+from ..swin.modeling_swin import SwinDropPath
+from .configuration_dinov3_vit import DINOv3ViTConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+@auto_docstring(
+    custom_intro="""
+    Output type of [`DINOv3ViTBackbone`], extending [`BackboneOutput`] with optional CLS tokens from
+    each selected feature stage (used when `config.return_class_token=True`).
+    """
+)
+@dataclass
+class DINOv3ViTBackboneOutput(BackboneOutput):
+    r"""
+    cls_tokens (`tuple(torch.FloatTensor)`, *optional*):
+        CLS token from each selected feature stage, each of shape `(batch_size, hidden_size)`.
+        Only present when `config.return_class_token=True`.
+    """
+
+    cls_tokens: tuple[torch.FloatTensor] | None = None
+
+
+class DINOv3ViTEmbeddings(nn.Module):
+    """
+    Construct the CLS token, mask token, position and patch embeddings.
+    """
+
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__()
+        self.config = config
+        self.cls_token = nn.Parameter(torch.randn(1, 1, config.hidden_size))
+        self.mask_token = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
+        self.register_tokens = nn.Parameter(torch.empty(1, config.num_register_tokens, config.hidden_size))
+        self.patch_embeddings = nn.Conv2d(
+            config.num_channels, config.hidden_size, kernel_size=config.patch_size, stride=config.patch_size
+        )
+
+    def forward(self, pixel_values: torch.Tensor, bool_masked_pos: torch.Tensor | None = None) -> torch.Tensor:
+        batch_size = pixel_values.shape[0]
+        target_dtype = self.patch_embeddings.weight.dtype
+
+        # (batch_size, num_channels, height, width) -> (batch_size, num_patches, hidden_size)
+        patch_embeddings = self.patch_embeddings(pixel_values.to(dtype=target_dtype))
+        patch_embeddings = patch_embeddings.flatten(2).transpose(1, 2)
+
+        if bool_masked_pos is not None:
+            mask_token = self.mask_token.to(patch_embeddings.dtype)
+            patch_embeddings = torch.where(bool_masked_pos.unsqueeze(-1), mask_token, patch_embeddings)
+
+        # Add CLS and register tokens
+        cls_token = self.cls_token.expand(batch_size, -1, -1)
+        register_tokens = self.register_tokens.expand(batch_size, -1, -1)
+        embeddings = torch.cat([cls_token, register_tokens, patch_embeddings], dim=1)
+
+        return embeddings
+
+
+@compile_compatible_method_lru_cache(maxsize=32)
+def get_patches_center_coordinates(
+    num_patches_h: int, num_patches_w: int, dtype: torch.dtype, device: torch.device
+) -> torch.Tensor:
+    """
+    Computes the 2D coordinates of the centers of image patches, normalized to the range [-1, +1].
+    The center of each patch is exactly halfway between its top-left and bottom-right corners.
+
+    Args:
+        num_patches_h (int): Number of patches along the vertical (height) axis.
+        num_patches_w (int): Number of patches along the horizontal (width) axis.
+        dtype (torch.dtype): The desired data type of the returned tensor.
+
+    Returns:
+        torch.Tensor: A tensor of shape (height * width, 2), where each row contains the (y, x)
+            coordinates of a patch center, normalized to [-1, +1].
+    """
+    coords_h = torch.arange(0.5, num_patches_h, dtype=dtype, device=device)
+    coords_w = torch.arange(0.5, num_patches_w, dtype=dtype, device=device)
+    coords_h = coords_h / num_patches_h
+    coords_w = coords_w / num_patches_w
+    # (height, width, 2) -> (height * width, 2)
+    coords = torch.stack(torch.meshgrid(coords_h, coords_w, indexing="ij"), dim=-1)
+    coords = coords.flatten(0, 1)
+    # Shift range [0, 1] to [-1, +1]
+    coords = 2.0 * coords - 1.0
+    return coords
+
+
+def augment_patches_center_coordinates(
+    coords: torch.Tensor,
+    shift: float | None = None,
+    jitter: float | None = None,
+    rescale: float | None = None,
+) -> torch.Tensor:
+    # Shift coords by adding a uniform value in [-shift, shift]
+    if shift is not None:
+        shift_hw = torch.empty((1, 2), device=coords.device, dtype=coords.dtype)
+        shift_hw = shift_hw.uniform_(-shift, shift)
+        coords = coords + shift_hw
+
+    # Jitter coords by multiplying the range [-1, 1] by a log-uniform value in [1/jitter, jitter]
+    if jitter is not None:
+        jitter_range = np.log(jitter)
+        jitter_hw = torch.empty((1, 2), device=coords.device, dtype=coords.dtype)
+        jitter_hw = jitter_hw.uniform_(-jitter_range, jitter_range).exp()
+        coords = coords * jitter_hw
+
+    # Rescale coords by multiplying the range [-1, 1] by a log-uniform value in [1/rescale, rescale]
+    if rescale is not None:
+        rescale_range = np.log(rescale)
+        rescale_hw = torch.empty(1, device=coords.device, dtype=coords.dtype)
+        rescale_hw = rescale_hw.uniform_(-rescale_range, rescale_range).exp()
+        coords = coords * rescale_hw
+
+    return coords
+
+
+class DINOv3ViTRopePositionEmbedding(nn.Module):
+    inv_freq: torch.Tensor
+
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__()
+
+        self.config = config
+        self.base = config.rope_theta
+        self.head_dim = config.hidden_size // config.num_attention_heads
+        self.num_patches_h = config.image_size // config.patch_size
+        self.num_patches_w = config.image_size // config.patch_size
+
+        inv_freq = 1 / self.base ** torch.arange(0, 1, 4 / self.head_dim, dtype=torch.float32)  # (head_dim / 4,)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+    def forward(self, pixel_values: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
+        _, _, height, width = pixel_values.shape
+        num_patches_h = height // self.config.patch_size
+        num_patches_w = width // self.config.patch_size
+
+        device = pixel_values.device
+        device_type = device.type if isinstance(device.type, str) and device.type != "mps" else "cpu"
+
+        with maybe_autocast(device_type=device_type, enabled=False):  # Force float32
+            # Although we could precompute static patch_coords from image_size and patch_size in the config,
+            # the model was trained with random_scale, so it can process images of varying sizes.
+            # Therefore, it's better to compute patch_coords dynamically (with lru_cache).
+            patch_coords = get_patches_center_coordinates(
+                num_patches_h, num_patches_w, dtype=torch.float32, device=device
+            )
+            if self.training:
+                patch_coords = augment_patches_center_coordinates(
+                    patch_coords,
+                    shift=self.config.pos_embed_shift,
+                    jitter=self.config.pos_embed_jitter,
+                    rescale=self.config.pos_embed_rescale,
+                )
+
+            # (height * width, 2, head_dim / 4) -> (height * width, head_dim / 2) -> (height * width, head_dim)
+            angles = 2 * math.pi * patch_coords[:, :, None] * self.inv_freq[None, None, :]
+            angles = angles.flatten(1, 2)
+            angles = angles.tile(2)
+
+            cos = torch.cos(angles)
+            sin = torch.sin(angles)
+
+        dtype = pixel_values.dtype
+        return cos.to(dtype=dtype), sin.to(dtype=dtype)
+
+
+def apply_rotary_pos_emb(
+    q: torch.Tensor, k: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor, **kwargs
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """Applies Rotary Position Embedding to the query and key tensors, but only to the patch tokens,
+    ignoring the prefix tokens (cls token and register tokens).
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+
+    num_tokens = q.shape[-2]
+    num_patches = sin.shape[-2]
+    num_prefix_tokens = num_tokens - num_patches  # cls token + register tokens
+
+    q_prefix_tokens, q_patches = q.split((num_prefix_tokens, num_patches), dim=-2)
+    k_prefix_tokens, k_patches = k.split((num_prefix_tokens, num_patches), dim=-2)
+
+    # apply rope only to patch tokens
+    q_patches = (q_patches * cos) + (rotate_half(q_patches) * sin)
+    k_patches = (k_patches * cos) + (rotate_half(k_patches) * sin)
+
+    q = torch.cat((q_prefix_tokens, q_patches), dim=-2)
+    k = torch.cat((k_prefix_tokens, k_patches), dim=-2)
+
+    return q, k
+
+
+class DINOv3ViTAttention(PixtralAttention):
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__(config)
+
+        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.query_bias)
+        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.key_bias)
+        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.value_bias)
+        self.o_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.proj_bias)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple[torch.Tensor, torch.Tensor | None]:
+        """Input shape: Batch x Time x Channel"""
+
+        batch_size, patches, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(batch_size, patches, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, attn_weights
+
+
+class DINOv3ViTLayerScale(Dinov2LayerScale):
+    pass
+
+
+class DINOv3ViTMLP(ArceeMLP):
+    pass
+
+
+class DINOv3ViTGatedMLP(LlamaMLP):
+    pass
+
+
+class Dinov3ViTDropPath(SwinDropPath):
+    pass
+
+
+class DINOv3ViTLayer(GradientCheckpointingLayer):
+    """This corresponds to the Block class in the original implementation."""
+
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__()
+
+        self.norm1 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.attention = DINOv3ViTAttention(config)
+        self.layer_scale1 = DINOv3ViTLayerScale(config)
+        self.drop_path = Dinov3ViTDropPath(config.drop_path_rate) if config.drop_path_rate > 0.0 else nn.Identity()
+
+        self.norm2 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+        if config.use_gated_mlp:
+            self.mlp = DINOv3ViTGatedMLP(config)
+        else:
+            self.mlp = DINOv3ViTMLP(config)
+        self.layer_scale2 = DINOv3ViTLayerScale(config)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.Tensor:
+        # Attention with residual connection
+        residual = hidden_states
+        hidden_states = self.norm1(hidden_states)
+        hidden_states, _ = self.attention(
+            hidden_states,
+            attention_mask=attention_mask,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = self.layer_scale1(hidden_states)
+        hidden_states = self.drop_path(hidden_states) + residual
+
+        # MLP with residual connection
+        residual = hidden_states
+        hidden_states = self.norm2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = self.layer_scale2(hidden_states)
+        hidden_states = self.drop_path(hidden_states) + residual
+
+        return hidden_states
+
+
+@auto_docstring
+class DINOv3ViTPreTrainedModel(Dinov2PreTrainedModel):
+    base_model_prefix = "model"
+    _can_record_outputs = {
+        "hidden_states": DINOv3ViTLayer,
+        "attentions": DINOv3ViTAttention,
+    }
+
+    @torch.no_grad()
+    def _init_weights(self, module) -> None:
+        """Initialize the weights"""
+        if isinstance(module, (nn.Linear, nn.Conv2d)):
+            init.trunc_normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                init.zeros_(module.bias)
+        elif isinstance(module, nn.LayerNorm):
+            init.zeros_(module.bias)
+            init.ones_(module.weight)
+        elif isinstance(module, DINOv3ViTEmbeddings):
+            init.trunc_normal_(module.cls_token, mean=0.0, std=self.config.initializer_range)
+            if module.config.num_register_tokens > 0:
+                init.trunc_normal_(module.register_tokens, mean=0.0, std=self.config.initializer_range)
+            init.zeros_(module.mask_token)
+        elif isinstance(module, DINOv3ViTLayerScale):
+            init.constant_(module.lambda1, self.config.layerscale_value)
+        elif isinstance(module, DINOv3ViTRopePositionEmbedding):
+            inv_freq = 1 / module.base ** torch.arange(0, 1, 4 / module.head_dim, dtype=torch.float32)
+            init.copy_(module.inv_freq, inv_freq)
+
+
+class DINOv3ViTEncoder(DINOv3ViTPreTrainedModel):
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__(config)
+        self.layer = nn.ModuleList([DINOv3ViTLayer(config) for _ in range(config.num_hidden_layers)])
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @merge_with_config_defaults
+    @capture_outputs(tie_last_hidden_states=False)
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutput:
+        for layer_module in self.layer:
+            hidden_states = layer_module(hidden_states, position_embeddings=position_embeddings, **kwargs)
+
+        return BaseModelOutput(last_hidden_state=hidden_states)
+
+
+@auto_docstring
+class DINOv3ViTModel(DINOv3ViTPreTrainedModel):
+    def __init__(self, config: DINOv3ViTConfig):
+        super().__init__(config)
+        self.embeddings = DINOv3ViTEmbeddings(config)
+        self.rope_embeddings = DINOv3ViTRopePositionEmbedding(config)
+        self.model = DINOv3ViTEncoder(config)
+        self.norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings.patch_embeddings
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        pixel_values: torch.Tensor,
+        bool_masked_pos: torch.Tensor | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutputWithPooling:
+        r"""
+        bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, sequence_length)`):
+            Boolean masked positions. Indicates which patches are masked (1) and which aren't (0). Only relevant for
+            pre-training.
+        """
+
+        pixel_values = pixel_values.to(self.embeddings.patch_embeddings.weight.dtype)
+        hidden_states = self.embeddings(pixel_values, bool_masked_pos=bool_masked_pos)
+        position_embeddings = self.rope_embeddings(pixel_values)
+
+        output = self.model(hidden_states, position_embeddings, **kwargs)
+        sequence_output = self.norm(output.last_hidden_state)
+        pooled_output = sequence_output[:, 0, :]
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            hidden_states=output.hidden_states,
+            attentions=output.attentions,
+        )
+
+
+@auto_docstring
+class DINOv3ViTBackbone(BackboneMixin, DINOv3ViTPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embeddings = DINOv3ViTEmbeddings(config)
+        self.rope_embeddings = DINOv3ViTRopePositionEmbedding(config)
+        self.model = DINOv3ViTEncoder(config)
+        self.norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.gradient_checkpointing = False
+
+        self.num_features = [config.hidden_size for _ in range(config.num_hidden_layers + 1)]
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings.patch_embeddings
+
+    @can_return_tuple
+    @filter_output_hidden_states
+    @auto_docstring
+    def forward(
+        self,
+        pixel_values: torch.Tensor,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> DINOv3ViTBackboneOutput:
+        pixel_values = pixel_values.to(self.embeddings.patch_embeddings.weight.dtype)
+        hidden_states = self.embeddings(pixel_values)
+        position_embeddings = self.rope_embeddings(pixel_values)
+
+        kwargs["output_hidden_states"] = True  # required to extract layers for the stages
+        output = self.model(hidden_states, position_embeddings, **kwargs)
+        stage_hidden_states = output.hidden_states
+
+        batch_size, _, image_height, image_width = pixel_values.shape
+        patch_size = self.config.patch_size
+        num_patches_height = image_height // patch_size
+        num_patches_width = image_width // patch_size
+
+        num_prefix = 1 + getattr(self.config, "num_register_tokens", 0)
+        return_class_token = getattr(self.config, "return_class_token", False)
+
+        feature_maps, cls_tokens = [], []
+        sequence_output = None
+        last_stage_idx = len(self.stage_names) - 1
+        for idx, (stage_name, hidden_state) in enumerate(zip(self.stage_names, stage_hidden_states)):
+            if idx == last_stage_idx:
+                hidden_state = self.norm(hidden_state)
+                sequence_output = hidden_state
+            elif self.config.apply_layernorm:
+                hidden_state = self.norm(hidden_state)
+
+            if stage_name in self.out_features:
+                if return_class_token:
+                    cls_tokens.append(hidden_state[:, 0, :])
+                patch_tokens = hidden_state[:, num_prefix:, :]
+                if self.config.reshape_hidden_states:
+                    fmap = (
+                        patch_tokens.reshape(batch_size, num_patches_height, num_patches_width, patch_tokens.shape[-1])
+                        .permute(0, 3, 1, 2)
+                        .contiguous()
+                    )
+                else:
+                    fmap = patch_tokens
+
+                feature_maps.append(fmap)
+
+        output = DINOv3ViTBackboneOutput(
+            feature_maps=tuple(feature_maps),
+            cls_tokens=tuple(cls_tokens) if return_class_token else None,
+            hidden_states=output.hidden_states,
+            attentions=output.attentions,
+        )
+        output.last_hidden_state = sequence_output
+
+        return output
+
+
+__all__ = ["DINOv3ViTModel", "DINOv3ViTPreTrainedModel", "DINOv3ViTBackbone"]

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/lfm2_moe/configuration_lfm2_moe.py

@@ -0,0 +1,84 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from huggingface_hub.dataclasses import strict
+
+from ...configuration_utils import PreTrainedConfig
+from ...utils import auto_docstring
+
+
+@auto_docstring(checkpoint="LiquidAI/LFM2-8B-A1B")
+@strict
+class Lfm2MoeConfig(PreTrainedConfig):
+    r"""
+    conv_bias (`bool`, *optional*, defaults to `False`):
+        Whether to use bias in the conv layers.
+    conv_L_cache (`int`, *optional*, defaults to 3):
+        L_cache dim in the conv layers.
+    num_dense_layers (`int`, *optional*, defaults to 2):
+        Number of dense Lfm2MoeMLP layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
+    use_expert_bias (`bool`, *optional*, defaults to `True`):
+        Whether to use the expert bias on the routing weights.
+
+    ```python
+    >>> from transformers import Lfm2MoeModel, Lfm2MoeConfig
+
+    >>> # Initializing a LFM2 Moe model
+    >>> configuration = Lfm2MoeConfig()
+
+    >>> # Initializing a model from the LFM2-8B-A1B style configuration
+    >>> model = Lfm2MoeModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```
+    """
+
+    model_type = "lfm2_moe"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    default_theta = 1000000.0
+
+    vocab_size: int = 65536
+    hidden_size: int = 2048
+    intermediate_size: int = 7168
+    moe_intermediate_size: int = 1792
+    num_hidden_layers: int = 32
+    pad_token_id: int | None = 0
+    bos_token_id: int | None = 1
+    eos_token_id: int | list[int] | None = 2
+    tie_word_embeddings: bool = True
+    rope_parameters: dict | None = None
+    max_position_embeddings: int = 128_000
+    initializer_range: float = 0.02
+    use_cache: bool = True
+    norm_eps: float = 0.00001
+    num_attention_heads: int = 32
+    num_key_value_heads: int = 8
+    conv_bias: bool = False
+    conv_L_cache: int = 3
+    num_dense_layers: int = 2
+    num_experts_per_tok: int = 4
+    num_experts: int = 32
+    use_expert_bias: bool = True
+    routed_scaling_factor: float = 1.0
+    norm_topk_prob: bool = True
+    layer_types: list[str] | None = None
+
+    def __post_init__(self, **kwargs):
+        self.tie_word_embeddings = kwargs.pop("tie_embedding", self.tie_word_embeddings)
+        super().__post_init__(**kwargs)
+
+
+__all__ = ["Lfm2MoeConfig"]

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/lfm2_moe/modular_lfm2_moe.py

@@ -0,0 +1,214 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+
+from ... import initialization as init
+from ...cache_utils import Cache, DynamicCache
+from ...masking_utils import create_causal_mask
+from ...modeling_outputs import MoeModelOutputWithPast
+from ...modeling_utils import PreTrainedModel
+from ...processing_utils import Unpack
+from ...utils import TransformersKwargs, logging
+from ...utils.import_utils import is_causal_conv1d_available
+from ..lfm2.modeling_lfm2 import (
+    Lfm2Attention,
+    Lfm2DecoderLayer,
+    Lfm2MLP,
+    Lfm2RotaryEmbedding,
+    Lfm2ShortConv,
+)
+from ..llama.modeling_llama import LlamaForCausalLM, LlamaPreTrainedModel, LlamaRMSNorm
+from ..mixtral.modeling_mixtral import MixtralModel
+from ..qwen2_moe.modeling_qwen2_moe import Qwen2MoeExperts
+from .configuration_lfm2_moe import Lfm2MoeConfig
+
+
+if is_causal_conv1d_available():
+    from causal_conv1d import causal_conv1d_fn, causal_conv1d_update
+else:
+    causal_conv1d_fn, causal_conv1d_update = None, None
+
+
+kernel_modules = (causal_conv1d_fn, causal_conv1d_update)
+is_fast_path_available = all(kernel_modules)
+
+
+logger = logging.get_logger(__name__)
+
+
+class Lfm2MoeRMSNorm(LlamaRMSNorm):
+    pass
+
+
+class Lfm2MoeRotaryEmbedding(Lfm2RotaryEmbedding):
+    pass
+
+
+class Lfm2MoeMLP(Lfm2MLP):
+    def __init__(self, config: Lfm2MoeConfig, intermediate_size: int | None = None):
+        nn.Module.__init__(self)
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size if intermediate_size is None else intermediate_size
+        self.w1 = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.w3 = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.w2 = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+
+
+class Lfm2MoeExperts(Qwen2MoeExperts):
+    def __init__(self, config):
+        super().__init__(config)
+        self.act_fn = F.silu
+
+
+class Lfm2MoeSparseMoeBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.top_k = config.num_experts_per_tok
+        self.routed_scaling_factor = config.routed_scaling_factor
+        self.norm_topk_prob = config.norm_topk_prob
+        self.use_expert_bias = config.use_expert_bias
+
+        self.gate = nn.Linear(config.hidden_size, config.num_experts, bias=False)
+        self.experts = Lfm2MoeExperts(config)
+        if self.use_expert_bias:
+            self.register_buffer("expert_bias", torch.zeros(config.num_experts, dtype=torch.float32))
+
+    def route_tokens_to_experts(self, router_logits):
+        routing_weights = router_logits.sigmoid()
+        if self.use_expert_bias:
+            scores_for_routing = routing_weights + self.expert_bias
+            _, selected_experts = torch.topk(scores_for_routing, k=self.top_k, dim=-1)
+            routing_weights = torch.gather(routing_weights, dim=1, index=selected_experts).type_as(router_logits)
+        else:
+            routing_weights, selected_experts = torch.topk(routing_weights, k=self.top_k, dim=-1)
+
+        if self.norm_topk_prob:
+            routing_weights = routing_weights / (routing_weights.sum(dim=-1, keepdim=True) + 1e-6)
+        routing_weights = routing_weights * self.routed_scaling_factor
+        return selected_experts, routing_weights
+
+    def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        hidden_states_reshaped = hidden_states.view(-1, hidden_dim)
+        router_logits = self.gate(hidden_states_reshaped)
+        selected_experts, routing_weights = self.route_tokens_to_experts(router_logits)
+        final_hidden_states = self.experts(hidden_states_reshaped, selected_experts, routing_weights)
+        return final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
+
+
+class Lfm2MoeAttention(Lfm2Attention):
+    pass
+
+
+class Lfm2MoeShortConv(Lfm2ShortConv):
+    pass
+
+
+class Lfm2MoeDecoderLayer(Lfm2DecoderLayer):
+    def __init__(self, config: Lfm2MoeConfig, layer_idx: int):
+        super().__init__(config, layer_idx)
+        self.feed_forward = (
+            Lfm2MoeMLP(config, intermediate_size=config.intermediate_size)
+            if layer_idx < config.num_dense_layers
+            else Lfm2MoeSparseMoeBlock(config)
+        )
+
+
+class Lfm2MoePreTrainedModel(LlamaPreTrainedModel):
+    _can_compile_fullgraph = False  # uses a non-compilable cache class
+
+    @torch.no_grad()
+    def _init_weights(self, module):
+        PreTrainedModel._init_weights(self, module)
+        if isinstance(module, Lfm2MoeExperts):
+            init.normal_(module.gate_up_proj, mean=0.0, std=self.config.initializer_range)
+            init.normal_(module.down_proj, mean=0.0, std=self.config.initializer_range)
+        elif isinstance(module, Lfm2MoeSparseMoeBlock):
+            if module.use_expert_bias:
+                init.zeros_(module.expert_bias)
+
+
+class Lfm2MoeModel(MixtralModel):
+    def __init__(self, config: Lfm2MoeConfig):
+        super().__init__(config)
+        self.pos_emb = Lfm2MoeRotaryEmbedding(config)
+        self.embedding_norm = Lfm2MoeRMSNorm(config.hidden_size, eps=config.norm_eps)
+        del self.norm
+        del self.rotary_emb
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        use_cache: bool | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> MoeModelOutputWithPast:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache(config=self.config)
+
+        if position_ids is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            position_ids = torch.arange(inputs_embeds.shape[1], device=inputs_embeds.device) + past_seen_tokens
+            position_ids = position_ids.unsqueeze(0)
+
+        causal_mask = create_causal_mask(
+            config=self.config,
+            inputs_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+        # Skip masking for decoding stage. We check shape here to be compile-friendly
+        linear_attention = attention_mask if inputs_embeds.shape[1] != 1 else None
+
+        hidden_states = inputs_embeds
+        position_embeddings = self.pos_emb(hidden_states, position_ids=position_ids)
+
+        # decoder layers
+        for i, decoder_layer in enumerate(self.layers[: self.config.num_hidden_layers]):
+            layer_mask = causal_mask if self.config.layer_types[i] == "full_attention" else linear_attention
+            hidden_states = decoder_layer(
+                hidden_states,
+                attention_mask=layer_mask,
+                position_ids=position_ids,
+                past_key_values=past_key_values,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+
+        hidden_states = self.embedding_norm(hidden_states)
+
+        return MoeModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+class Lfm2MoeForCausalLM(LlamaForCausalLM):
+    pass
+
+
+__all__ = ["Lfm2MoeForCausalLM", "Lfm2MoeModel", "Lfm2MoePreTrainedModel"]