gpt-5.6-sol

Script running with cell ID 151
Wall time 11.0 seconds
Output:

Script completed
Wall time 41.8 seconds
Output:
{"chunk_id":"2502b2","wall_time_seconds":60.001588063,"session_id":3244,"original_token_count":0,"output":""}

Script running with cell ID 152
Wall time 11.0 seconds
Output:

Script completed
Wall time 43.6 seconds
Output:
{"chunk_id":"b26881","wall_time_seconds":60.002524569,"session_id":3244,"original_token_count":33,"output":"\rProcessed prompts:  38%|███▊      | 36/96 [07:31<11:19, 11.33s/it, est. speed input: 10.47 toks/s, output: 693.92 toks/s]"}

Script running with cell ID 153
Wall time 11.0 seconds
Output:

Script completed
Wall time 36.6 seconds
Output:
{"chunk_id":"cc8cb2","wall_time_seconds":60.001474891,"session_id":3244,"original_token_count":0,"output":""}

Script running with cell ID 154
Wall time 11.0 seconds
Output:

Script completed
Wall time 41.6 seconds
Output:
{"chunk_id":"f402f7","wall_time_seconds":60.001311744,"session_id":3244,"original_token_count":163,"output":"\rProcessed prompts:  29%|██▊       | 24/84 [10:00<24:12, 24.21s/it, est. speed input: 3.00 toks/s, output: 477.45 toks/s]\rProcessed prompts:  50%|█████     | 48/96 [10:26<10:04, 12.60s/it, est. speed input: 9.01 toks/s, output: 793.74 toks/s] \rProcessed prompts:  14%|█▍        | 12/84 [10:26<1:02:41, 52.25s/it, est. speed input: 1.61 toks/s, output: 282.58 toks/s]\rProcessed prompts:  62%|██████▎   | 60/96 [10:46<05:11,  8.66s/it, est. speed input: 10.97 toks/s, output: 1082.38 toks/s]\rProcessed prompts:  25%|██▌       | 24/96 [11:01<32:06, 26.76s/it, est. speed input: 4.81 toks/s, output: 521.49 toks/s]"}

Script running with cell ID 155
Wall time 11.0 seconds
Output:

Script completed
Wall time 43.6 seconds
Output:
{"chunk_id":"39cd95","wall_time_seconds":60.001383512,"session_id":3244,"original_token_count":68,"output":"\rProcessed prompts:  75%|███████▌  | 72/96 [11:17<02:38,  6.59s/it, est. speed input: 12.03 toks/s, output: 1341.83 toks/s]\rProcessed prompts:  43%|████▎     | 36/84 [12:01<14:13, 17.77s/it, est. speed input: 3.87 toks/s, output: 656.70 toks/s]"}

Script running with cell ID 156
Wall time 11.0 seconds
Output:

Script completed
Wall time 38.6 seconds
Output:
Warning: truncated output (original token count: 276)
Total output lines: 1

{"chunk_id":"74247b","wall_time_seconds":60.000855985,"session_id":3244,"original_token_count":246,"output":"\rProcessed prompts:  57%|█████▋    | 48/84 [12:13<06:41, 11.14s/it, est. speed input: 6.02 toks/s, output: 1030.19 toks/s]\rProcessed prompts:  88%|████████▊ | 84/96 [12:28<01:16,  6.38s/it, est. speed input: 17.56 toks/s, output: 1669.78 toks/s]\rProcessed prompts:  71%|███████▏  | 60/84 [12:45<03:13,  8.08s/it, est. speed input: 7.40 toks/s…26 tokens truncated…5.40s/it, est. speed input: 8.85 toks/s, output: 1766.51 toks/s]\rProcessed prompts: 100%|██████████| 96/96 [13:06<00:00,  5.36s/it, est. speed input: 19.30 toks/s, output: 2109.40 toks/s]\rProcessed prompts: 100%|██████████| 96/96 [13:06<00:00,  5.36s/it, est. speed input: 19.30 toks/s, output: 2109.40 toks/s]\rProcessed prompts: 100%|██████████| 96/96 [13:06<00:00,  8.19s/it, est. speed input: 19.30 toks/s, output: 2109.40 toks/s]\n"}

Script running with cell ID 157
Wall time 11.0 seconds
Output:

Script completed
Wall time 43.0 seconds
Output:
Warning: truncated output (original token count: 306)
Total output lines: 1

{"chunk_id":"ca8ad8","wall_time_seconds":60.002537519,"session_id":3244,"original_token_count":306,"output":"Warning: truncated output (original token count: 306)\nTotal output lines: 2\n\n\rProcessed prompts: 100%|██████████| 84/84 [13:25<00:00,  4.66s/it, est. speed input: 10.84 toks/s, output: 2045.79 toks/s]\rProcessed prompts: 100%|██████████| 84/84 [13:25<00:00,  4.66s/it, est. speed input: 10.84 toks/s, output: 2045.79 toks/s]\rProcessed prompts: 10…56 tokens truncated…     | 36/84 [13:47<13:43, 17.15s/it, est. speed input: 4.20 toks/s, output: 870.72 toks/s]\rProcessed prompts:  50%|█████     | 48/96 [13:56<10:28, 13.09s/it, est. speed input: 7.92 toks/s, output: 1079.32 toks/s]\rProcessed prompts:  57%|█████▋    | 48/84 [13:58<06:26, 10.73s/it, est. speed input: 5.87 toks/s, output: 1197.87 toks/s]\rProcessed prompts:  71%|███████▏  | 60/84 [14:22<03:02,  7.60s/it, est. speed input: 7.09 toks/s, output: 1494.57 toks/s]"}

Script running with cell ID 158
Wall time 11.0 seconds
Output:

Script completed
Wall time 38.1 seconds
Output:
{"chunk_id":"fe06b5","wall_time_seconds":60.002059484,"session_id":3244,"original_token_count":177,"output":"\rProcessed prompts:  86%|████████▌ | 72/84 [14:49<01:09,  5.78s/it, est. speed input: 8.21 toks/s, output: 1739.12 toks/s]\rProcessed prompts: 100%|██████████| 84/84 [15:17<00:00,  4.64s/it, est. speed input: 10.84 toks/s, output: 2048.83 toks/s]\rProcessed prompts: 100%|██████████| 84/84 [15:17<00:00,  4.64s/it, est. speed input: 10.84 toks/s, output: 2048.83 toks/s]\rProcessed prompts: 100%|██████████| 84/84 [15:17<00:00, 10.92s/it, est. speed input: 10.84 toks/s, output: 2048.83 toks/s]\n\rProcessed prompts:  62%|██████▎   | 60/96 [15:21<06:33, 10.93s/it, est. speed input: 8.88 toks/s, output: 1224.45 toks/s]"}

Script running with cell ID 159
Wall time 11.0 seconds
Output:

Script completed
Wall time 32.8 seconds
Output:
{"chunk_id":"3974fc","wall_time_seconds":60.001815364,"session_id":3244,"original_token_count":213,"output":"\rProcessed prompts:  75%|███████▌  | 72/96 [16:27<03:38,  9.09s/it, est. speed input: 9.79 toks/s, output: 1389.28 toks/s]\rProcessed prompts:  88%|████████▊ | 84/96 [16:48<01:20,  6.70s/it, est. speed input: 10.88 toks/s, output: 1724.54 toks/s]\rProcessed prompts: 100%|██████████| 96/96 [16:50<00:00,  4.60s/it, est. speed input: 12.95 toks/s, output: 2093.39 toks/s]\rProcessed prompts: 100%|██████████| 96/96 [16:50<00:00,  4.60s/it, est. speed input: 12.95 toks/s, output: 2093.39 toks/s]\rProcessed prompts: 100%|██████████| 96/96 [16:50<00:00, 10.52s/it, est. speed input: 12.95 toks/s, output: 2093.39 toks/s]\n\nFINAL avg@12 accuracy: 56.11% (202/360 solutions correct on 30 problems)\nResults written to /workspace/eval_results/dev_eval_1785739883.json\n"}

Script completed
Wall time 0.1 seconds
Output:
12516
 README.md       |  4 +---
 opsd_trainer.py | 24 ++++++++++++++++++------
 2 files changed, 19 insertions(+), 9 deletions(-)
diff --git a/opsd_trainer.py b/opsd_trainer.py
index c151bd7..7549be4 100644
--- a/opsd_trainer.py
+++ b/opsd_trainer.py
@@ -463,20 +463,32 @@ class OPSDTrainer(SFTTrainer):
         if token_clip is not None:
             jsd = jsd.clamp(max=token_clip)
 
-        # Masking
+        # Masking and trajectory credit assignment.  Early reasoning tokens are
+        # more likely to encode a consequential branch choice; late tokens are
+        # increasingly determined by the already-long prefix.  Give early
+        # positions moderately more credit without changing the total loss
+        # scale.  The linear 1.5 -> 0.5 schedule has mean one on every non-empty
+        # trajectory, so learning-rate and clipping behavior remain comparable
+        # to the original uniformly averaged OPSD objective.
         if labels is not None:
             mask = labels != -100
-            jsd = jsd[mask]
+            token_divergence = jsd.sum(dim=-1)
+            positions = mask.long().cumsum(dim=-1) - 1
+            lengths = mask.sum(dim=-1, keepdim=True).clamp_min(1)
+            progress = positions.to(token_divergence.dtype) / (lengths - 1).clamp_min(1)
+            credit = (1.5 - progress).clamp_min(0.5) * mask
 
         # Apply reduction
         if reduction == "batchmean":
-            return jsd.sum() / mask.sum() if labels is not None else jsd.sum() / jsd.size(0)
+            if labels is not None:
+                return (token_divergence * credit).sum() / mask.sum().clamp_min(1)
+            return jsd.sum() / jsd.size(0)
         elif reduction == "sum":
-            return jsd.sum()
+            return (token_divergence * credit).sum() if labels is not None else jsd.sum()
         elif reduction == "mean":
-            return jsd.mean()
+            return (token_divergence * credit).sum() / credit.sum().clamp_min(1) if labels is not None else jsd.mean()
         else:
-            return jsd
+            return token_divergence * credit if labels is not None else jsd
 
     def _update_ema(self):
         """Update EMA parameters after an optimizer step.
RECIPE
# OPSD method recipe — whitelisted hyper-parameter overrides (KEY=VALUE per line).
# Empty/all-commented == the OPSD baseline recipe on top of your /workspace/repo
# code changes. The scorer re-trains with THIS file + your repo for 100 steps.
# Budget knobs (max_steps, epochs, batch, gradient_accumulation, num_processes,
# model) are FIXED by the scorer and ignored here.
#
# learning_rate=5e-6
# lora_r=64
# lora_alpha=128
# beta=0
# jsd_token_clip=0.05
# temperature=1.1
# lmbda=1
# max_completion_length=1024        # <= 4096
# lr_scheduler_type=constant
# fixed_teacher=true
# use_ema_teacher=false
# reason_first=false
# teacher_thinking=false

Script completed
Wall time 1.2 seconds
Output:
{}{"chunk_id":"517f8e","wall_time_seconds":1.001851853,"session_id":53588,"original_token_count":0,"output":""}

Script running with cell ID 162
Wall time 11.0 seconds
Output:

Script completed
Wall time 43.1 seconds
Output:
Warning: truncated output (original token count: 279)
Total output lines: 1

{"chunk_id":"42395a","wall_time_seconds":60.001255324,"session_id":53588,"original_token_count":4092,"output":"Warning: truncated output (original token count: 4092)\nTotal output lines: 197\n\n<string>:24: FutureWarning: `torch_dtype` is deprecated and will be removed in version 0.27.0, please use `dtype` instead.\n\n================================================================================\nRUN CONFIGURATION\n===================…59 tokens truncated…, avg length: 735.2, speed: 902.8 tok/s\nvLLM generation done - elapsed time: 3.26s, prompts: 4, total tokens: 2933, avg length: 733.2, speed: 899.8 tok/s\nvLLM generation done - elapsed time: 3.27s, prompts: 4, total tokens: 3232, avg length: 808.0, speed: 987.4 tok/s\nvLLM generation done - elapsed time: 3.33s, prompts: 4, total tokens: 3575, avg length: 893.8, speed: 1073.3 tok/s\n\r  3%|▎         | 3/100 [00:23<12:50,  7.95s/it]"}