v6_sft_only_v2: training log

v6_sft_only_v2/training.log

@@ -0,0 +1,615 @@
+🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.
+🦥 Unsloth Zoo will now patch everything to make training faster!
+Loading unsloth/Qwen3-4B-Instruct-2507...
+==((====))==  Unsloth 2026.4.8: Fast Qwen3 patching. Transformers: 4.56.2. vLLM: 0.15.1.
+   \\   /|    NVIDIA L40S. Num GPUs = 1. Max memory: 44.392 GB. Platform: Linux.
+O^O/ \_/ \    Torch: 2.9.1+cu128. CUDA: 8.9. CUDA Toolkit: 12.8. Triton: 3.5.1
+\        /    Bfloat16 = TRUE. FA [Xformers = 0.0.33.post2. FA2 = False]
+ "-____-"     Free license: http://github.com/unslothai/unsloth
+Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!
+
+
+model.safetensors.index.json:   0%|          | 0.00/32.9k [00:00<?, ?B/s]
+model.safetensors.index.json: 100%|██████████| 32.9k/32.9k [00:00<00:00, 135MB/s]
+
+
+model-00001-of-00002.safetensors:   0%|          | 0.00/4.97G [00:00<?, ?B/s]
+
+model-00001-of-00002.safetensors:   1%|▏         | 67.1M/4.97G [00:01<01:19, 61.6MB/s]
+
+model-00001-of-00002.safetensors:  11%|█         | 536M/4.97G [00:02<00:17, 259MB/s]  
+
+model-00001-of-00002.safetensors:  62%|██████▏   | 3.08G/4.97G [00:03<00:01, 1.08GB/s]
+model-00001-of-00002.safetensors: 100%|██████████| 4.97G/4.97G [00:03<00:00, 1.29GB/s]
+
+
+model-00002-of-00002.safetensors:   0%|          | 0.00/3.08G [00:00<?, ?B/s]
+
+model-00002-of-00002.safetensors:   0%|          | 0.00/3.08G [00:01<?, ?B/s]
+
+model-00002-of-00002.safetensors:  50%|█████     | 1.54G/3.08G [00:02<00:01, 1.30GB/s]
+model-00002-of-00002.safetensors: 100%|██████████| 3.08G/3.08G [00:03<00:00, 1.02GB/s]
+
+
+Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]
+Loading checkpoint shards: 100%|██████████| 2/2 [00:00<00:00,  2.63it/s]
+
+
+generation_config.json:   0%|          | 0.00/237 [00:00<?, ?B/s]
+generation_config.json: 100%|██████████| 237/237 [00:00<00:00, 2.27MB/s]
+
+
+tokenizer_config.json:   0%|          | 0.00/9.65k [00:00<?, ?B/s]
+tokenizer_config.json: 100%|██████████| 9.65k/9.65k [00:00<00:00, 61.1MB/s]
+
+
+vocab.json:   0%|          | 0.00/2.78M [00:00<?, ?B/s]
+vocab.json: 100%|██████████| 2.78M/2.78M [00:00<00:00, 40.9MB/s]
+
+
+merges.txt:   0%|          | 0.00/1.67M [00:00<?, ?B/s]
+merges.txt: 100%|██████████| 1.67M/1.67M [00:00<00:00, 128MB/s]
+
+
+added_tokens.json:   0%|          | 0.00/707 [00:00<?, ?B/s]
+added_tokens.json: 100%|██████████| 707/707 [00:00<00:00, 6.15MB/s]
+
+
+special_tokens_map.json:   0%|          | 0.00/614 [00:00<?, ?B/s]
+special_tokens_map.json: 100%|██████████| 614/614 [00:00<00:00, 4.61MB/s]
+
+
+tokenizer.json:   0%|          | 0.00/11.4M [00:00<?, ?B/s]
+tokenizer.json: 100%|██████████| 11.4M/11.4M [00:00<00:00, 54.2MB/s]
+
+
+chat_template.jinja:   0%|          | 0.00/4.04k [00:00<?, ?B/s]
+chat_template.jinja: 100%|██████████| 4.04k/4.04k [00:00<00:00, 27.2MB/s]
+unsloth/Qwen3-4B-Instruct-2507 does not have a padding token! Will use pad_token = <|PAD_TOKEN|>.
+Unsloth 2026.4.8 patched 36 layers with 36 QKV layers, 36 O layers and 36 MLP layers.
+VRAM allocated: 8.20 GB
+
+══ SFT warm-start — sft_traces/merged_v6_aligned.jsonl ══
+  200 SFT examples loaded (chat format in `text`)
+
+
+Unsloth: Tokenizing ["text"] (num_proc=12):   0%|          | 0/200 [00:00<?, ? examples/s]
+
+Unsloth: Tokenizing ["text"] (num_proc=12):  26%|██▌       | 51/200 [00:01<00:03, 44.59 examples/s]
+Unsloth: Tokenizing ["text"] (num_proc=12): 100%|██████████| 200/200 [00:02<00:00, 95.33 examples/s]
+🦥 Unsloth: Padding-free auto-enabled, enabling faster training.
+==((====))==  Unsloth - 2x faster free finetuning | Num GPUs used = 1
+   \\   /|    Num examples = 200 | Num Epochs = 6 | Total steps = 150
+O^O/ \_/ \    Batch size per device = 2 | Gradient accumulation steps = 4
+\        /    Data Parallel GPUs = 1 | Total batch size (2 x 4 x 1) = 8
+ "-____-"     Trainable parameters = 33,030,144 of 4,055,498,240 (0.81% trained)
+
+
+  0%|          | 0/150 [00:00<?, ?it/s]Unsloth: Will smartly offload gradients to save VRAM!
+
+
+  1%|          | 1/150 [00:06<17:00,  6.85s/it]
+
+  1%|▏         | 2/150 [00:08<09:12,  3.73s/it]
+
+  2%|▏         | 3/150 [00:09<06:35,  2.69s/it]
+
+  3%|▎         | 4/150 [00:11<05:22,  2.21s/it]
+
+  3%|▎         | 5/150 [00:12<04:42,  1.95s/it]
+
+                                               
+{'loss': 3.2003, 'grad_norm': 2.315643072128296, 'learning_rate': 2.5e-05, 'epoch': 0.2}
+
+
+  3%|▎         | 5/150 [00:12<04:42,  1.95s/it]
+
+  4%|▍         | 6/150 [00:14<04:18,  1.79s/it]
+
+  5%|▍         | 7/150 [00:15<04:03,  1.70s/it]
+
+  5%|▌         | 8/150 [00:17<03:52,  1.64s/it]
+
+  6%|▌         | 9/150 [00:18<03:44,  1.59s/it]
+
+  7%|▋         | 10/150 [00:20<03:38,  1.56s/it]
+
+                                                
+{'loss': 2.9439, 'grad_norm': 1.7202048301696777, 'learning_rate': 4.9647887323943665e-05, 'epoch': 0.4}
+
+
+  7%|▋         | 10/150 [00:20<03:38,  1.56s/it]
+
+  7%|▋         | 11/150 [00:21<03:34,  1.54s/it]
+
+  8%|▊         | 12/150 [00:23<03:30,  1.52s/it]
+
+  9%|▊         | 13/150 [00:24<03:29,  1.53s/it]
+
+  9%|▉         | 14/150 [00:26<03:28,  1.53s/it]
+
+ 10%|█         | 15/150 [00:27<03:24,  1.51s/it]
+
+                                                
+{'loss': 2.4466, 'grad_norm': 0.7943887114524841, 'learning_rate': 4.788732394366197e-05, 'epoch': 0.6}
+
+
+ 10%|█         | 15/150 [00:27<03:24,  1.51s/it]
+
+ 11%|█         | 16/150 [00:29<03:23,  1.52s/it]
+
+ 11%|█▏        | 17/150 [00:30<03:20,  1.51s/it]
+
+ 12%|█▏        | 18/150 [00:32<03:19,  1.51s/it]
+
+ 13%|█▎        | 19/150 [00:33<03:15,  1.50s/it]
+
+ 13%|█▎        | 20/150 [00:35<03:15,  1.50s/it]
+
+                                                
+{'loss': 2.163, 'grad_norm': 0.7801544666290283, 'learning_rate': 4.6126760563380286e-05, 'epoch': 0.8}
+
+
+ 13%|█▎        | 20/150 [00:35<03:15,  1.50s/it]
+
+ 14%|█▍        | 21/150 [00:36<03:16,  1.52s/it]
+
+ 15%|█▍        | 22/150 [00:38<03:13,  1.51s/it]
+
+ 15%|█▌        | 23/150 [00:39<03:12,  1.52s/it]
+
+ 16%|█▌        | 24/150 [00:41<03:09,  1.50s/it]
+
+ 17%|█▋        | 25/150 [00:42<03:07,  1.50s/it]
+
+                                                
+{'loss': 1.9002, 'grad_norm': 0.780380129814148, 'learning_rate': 4.436619718309859e-05, 'epoch': 1.0}
+
+
+ 17%|█▋        | 25/150 [00:42<03:07,  1.50s/it]
+
+ 17%|█▋        | 26/150 [00:44<03:07,  1.52s/it]
+
+ 18%|█▊        | 27/150 [00:45<03:06,  1.51s/it]
+
+ 19%|█▊        | 28/150 [00:47<03:01,  1.49s/it]
+
+ 19%|█▉        | 29/150 [00:48<03:01,  1.50s/it]
+
+ 20%|██        | 30/150 [00:50<03:01,  1.51s/it]
+
+                                                
+{'loss': 1.6575, 'grad_norm': 0.7079782485961914, 'learning_rate': 4.26056338028169e-05, 'epoch': 1.2}
+
+
+ 20%|██        | 30/150 [00:50<03:01,  1.51s/it]
+
+ 21%|██        | 31/150 [00:51<02:56,  1.48s/it]
+
+ 21%|██▏       | 32/150 [00:53<02:54,  1.48s/it]
+
+ 22%|██▏       | 33/150 [00:54<02:54,  1.50s/it]
+
+ 23%|██▎       | 34/150 [00:56<02:54,  1.50s/it]
+
+ 23%|██▎       | 35/150 [00:57<02:54,  1.52s/it]
+
+                                                
+{'loss': 1.4154, 'grad_norm': 0.7800647616386414, 'learning_rate': 4.0845070422535214e-05, 'epoch': 1.4}
+
+
+ 23%|██▎       | 35/150 [00:57<02:54,  1.52s/it]
+
+ 24%|██▍       | 36/150 [00:59<02:54,  1.53s/it]
+
+ 25%|██▍       | 37/150 [01:01<02:52,  1.53s/it]
+
+ 25%|██▌       | 38/150 [01:02<02:49,  1.51s/it]
+
+ 26%|██▌       | 39/150 [01:03<02:47,  1.51s/it]
+
+ 27%|██▋       | 40/150 [01:05<02:48,  1.53s/it]
+
+                                                
+{'loss': 1.1887, 'grad_norm': 0.7795201539993286, 'learning_rate': 3.908450704225352e-05, 'epoch': 1.6}
+
+
+ 27%|██▋       | 40/150 [01:05<02:48,  1.53s/it]
+
+ 27%|██▋       | 41/150 [01:07<02:45,  1.52s/it]
+
+ 28%|██▊       | 42/150 [01:08<02:43,  1.52s/it]
+
+ 29%|██▊       | 43/150 [01:10<02:44,  1.54s/it]
+
+ 29%|██▉       | 44/150 [01:11<02:40,  1.51s/it]
+
+ 30%|███       | 45/150 [01:13<02:39,  1.52s/it]
+
+                                                
+{'loss': 0.9898, 'grad_norm': 0.5640354752540588, 'learning_rate': 3.7323943661971835e-05, 'epoch': 1.8}
+
+
+ 30%|███       | 45/150 [01:13<02:39,  1.52s/it]
+
+ 31%|███       | 46/150 [01:14<02:38,  1.52s/it]
+
+ 31%|███▏      | 47/150 [01:16<02:35,  1.51s/it]
+
+ 32%|███▏      | 48/150 [01:17<02:34,  1.52s/it]
+
+ 33%|███▎      | 49/150 [01:19<02:33,  1.52s/it]
+
+ 33%|███▎      | 50/150 [01:20<02:31,  1.51s/it]
+
+                                                
+{'loss': 0.9291, 'grad_norm': 0.43140193819999695, 'learning_rate': 3.556338028169014e-05, 'epoch': 2.0}
+
+
+ 33%|███▎      | 50/150 [01:20<02:31,  1.51s/it]
+
+ 34%|███▍      | 51/150 [01:22<02:30,  1.52s/it]
+
+ 35%|███▍      | 52/150 [01:23<02:29,  1.53s/it]
+
+ 35%|███▌      | 53/150 [01:25<02:27,  1.52s/it]
+
+ 36%|███▌      | 54/150 [01:26<02:23,  1.49s/it]
+
+ 37%|███▋      | 55/150 [01:28<02:21,  1.49s/it]
+
+                                                
+{'loss': 0.8827, 'grad_norm': 0.3559304475784302, 'learning_rate': 3.380281690140845e-05, 'epoch': 2.2}
+
+
+ 37%|███▋      | 55/150 [01:28<02:21,  1.49s/it]
+
+ 37%|███▋      | 56/150 [01:29<02:21,  1.51s/it]
+
+ 38%|███▊      | 57/150 [01:31<02:20,  1.51s/it]
+
+ 39%|███▊      | 58/150 [01:32<02:19,  1.52s/it]
+
+ 39%|███▉      | 59/150 [01:34<02:19,  1.53s/it]
+
+ 40%|████      | 60/150 [01:35<02:18,  1.54s/it]
+
+                                                
+{'loss': 0.8919, 'grad_norm': 0.3519206643104553, 'learning_rate': 3.204225352112676e-05, 'epoch': 2.4}
+
+
+ 40%|████      | 60/150 [01:35<02:18,  1.54s/it]
+
+ 41%|████      | 61/150 [01:37<02:16,  1.53s/it]
+
+ 41%|████▏     | 62/150 [01:38<02:14,  1.53s/it]
+
+ 42%|████▏     | 63/150 [01:40<02:12,  1.52s/it]
+
+ 43%|████▎     | 64/150 [01:42<02:11,  1.53s/it]
+
+ 43%|████▎     | 65/150 [01:43<02:09,  1.53s/it]
+
+                                                
+{'loss': 0.869, 'grad_norm': 0.3210051655769348, 'learning_rate': 3.028169014084507e-05, 'epoch': 2.6}
+
+
+ 43%|████▎     | 65/150 [01:43<02:09,  1.53s/it]
+
+ 44%|████▍     | 66/150 [01:45<02:08,  1.53s/it]
+
+ 45%|████▍     | 67/150 [01:46<02:05,  1.51s/it]
+
+ 45%|████▌     | 68/150 [01:48<02:03,  1.50s/it]
+
+ 46%|████▌     | 69/150 [01:49<02:02,  1.51s/it]
+
+ 47%|████▋     | 70/150 [01:51<02:02,  1.53s/it]
+
+                                                
+{'loss': 0.8251, 'grad_norm': 0.31960350275039673, 'learning_rate': 2.8521126760563384e-05, 'epoch': 2.8}
+
+
+ 47%|████▋     | 70/150 [01:51<02:02,  1.53s/it]
+
+ 47%|████▋     | 71/150 [01:52<02:01,  1.54s/it]
+
+ 48%|████▊     | 72/150 [01:54<01:59,  1.53s/it]
+
+ 49%|████▊     | 73/150 [01:55<01:58,  1.54s/it]
+
+ 49%|████▉     | 74/150 [01:57<01:56,  1.54s/it]
+
+ 50%|█████     | 75/150 [01:58<01:54,  1.53s/it]
+
+                                                
+{'loss': 0.8397, 'grad_norm': 0.3365177512168884, 'learning_rate': 2.676056338028169e-05, 'epoch': 3.0}
+
+
+ 50%|█████     | 75/150 [01:58<01:54,  1.53s/it]
+
+ 51%|█████     | 76/150 [02:00<01:52,  1.52s/it]
+
+ 51%|█████▏    | 77/150 [02:01<01:50,  1.52s/it]
+
+ 52%|█████▏    | 78/150 [02:03<01:50,  1.53s/it]
+
+ 53%|█████▎    | 79/150 [02:04<01:49,  1.54s/it]
+
+ 53%|█████▎    | 80/150 [02:06<01:48,  1.54s/it]
+
+                                                
+{'loss': 0.7999, 'grad_norm': 0.3121998608112335, 'learning_rate': 2.5e-05, 'epoch': 3.2}
+
+
+ 53%|█████▎    | 80/150 [02:06<01:48,  1.54s/it]
+
+ 54%|█████▍    | 81/150 [02:08<01:45,  1.53s/it]
+
+ 55%|█████▍    | 82/150 [02:09<01:43,  1.53s/it]
+
+ 55%|█████▌    | 83/150 [02:11<01:42,  1.52s/it]
+
+ 56%|█████▌    | 84/150 [02:12<01:42,  1.55s/it]
+
+ 57%|█████▋    | 85/150 [02:14<01:40,  1.54s/it]
+
+                                                
+{'loss': 0.8139, 'grad_norm': 0.33012324571609497, 'learning_rate': 2.323943661971831e-05, 'epoch': 3.4}
+
+
+ 57%|█████▋    | 85/150 [02:14<01:40,  1.54s/it]
+
+ 57%|█████▋    | 86/150 [02:15<01:37,  1.53s/it]
+
+ 58%|█████▊    | 87/150 [02:17<01:36,  1.54s/it]
+
+ 59%|█████▊    | 88/150 [02:18<01:34,  1.53s/it]
+
+ 59%|█████▉    | 89/150 [02:20<01:33,  1.53s/it]
+
+ 60%|██████    | 90/150 [02:21<01:31,  1.52s/it]
+
+                                                
+{'loss': 0.7965, 'grad_norm': 0.325542151927948, 'learning_rate': 2.147887323943662e-05, 'epoch': 3.6}
+
+
+ 60%|██████    | 90/150 [02:21<01:31,  1.52s/it]
+
+ 61%|██████    | 91/150 [02:23<01:29,  1.52s/it]
+
+ 61%|██████▏   | 92/150 [02:24<01:28,  1.53s/it]
+
+ 62%|██████▏   | 93/150 [02:26<01:27,  1.54s/it]
+
+ 63%|██████▎   | 94/150 [02:27<01:26,  1.54s/it]
+
+ 63%|██████▎   | 95/150 [02:29<01:24,  1.53s/it]
+
+                                                
+{'loss': 0.8084, 'grad_norm': 0.31672295928001404, 'learning_rate': 1.971830985915493e-05, 'epoch': 3.8}
+
+
+ 63%|██████▎   | 95/150 [02:29<01:24,  1.53s/it]
+
+ 64%|██████▍   | 96/150 [02:31<01:22,  1.54s/it]
+
+ 65%|██████▍   | 97/150 [02:32<01:21,  1.54s/it]
+
+ 65%|██████▌   | 98/150 [02:33<01:18,  1.51s/it]
+
+ 66%|██████▌   | 99/150 [02:35<01:17,  1.52s/it]
+
+ 67%|██████▋   | 100/150 [02:37<01:16,  1.53s/it]
+
+                                                 
+{'loss': 0.8154, 'grad_norm': 0.3328615128993988, 'learning_rate': 1.7957746478873243e-05, 'epoch': 4.0}
+
+
+ 67%|██████▋   | 100/150 [02:37<01:16,  1.53s/it]
+
+ 67%|██████▋   | 101/150 [02:38<01:15,  1.54s/it]
+
+ 68%|██████▊   | 102/150 [02:40<01:13,  1.54s/it]
+
+ 69%|██████▊   | 103/150 [02:41<01:12,  1.54s/it]
+
+ 69%|████���█▉   | 104/150 [02:43<01:11,  1.56s/it]
+
+ 70%|███████   | 105/150 [02:44<01:09,  1.55s/it]
+
+                                                 
+{'loss': 0.7967, 'grad_norm': 0.32036644220352173, 'learning_rate': 1.619718309859155e-05, 'epoch': 4.2}
+
+
+ 70%|███████   | 105/150 [02:44<01:09,  1.55s/it]
+
+ 71%|███████   | 106/150 [02:46<01:08,  1.55s/it]
+
+ 71%|███████▏  | 107/150 [02:47<01:06,  1.55s/it]
+
+ 72%|███████▏  | 108/150 [02:49<01:05,  1.56s/it]
+
+ 73%|███████▎  | 109/150 [02:51<01:03,  1.55s/it]
+
+ 73%|███████▎  | 110/150 [02:52<01:01,  1.53s/it]
+
+                                                 
+{'loss': 0.7825, 'grad_norm': 0.33183553814888, 'learning_rate': 1.443661971830986e-05, 'epoch': 4.4}
+
+
+ 73%|███████▎  | 110/150 [02:52<01:01,  1.53s/it]
+
+ 74%|███████▍  | 111/150 [02:54<00:59,  1.53s/it]
+
+ 75%|███████▍  | 112/150 [02:55<00:56,  1.50s/it]
+
+ 75%|███████▌  | 113/150 [02:57<00:55,  1.51s/it]
+
+ 76%|███████▌  | 114/150 [02:58<00:54,  1.52s/it]
+
+ 77%|███████▋  | 115/150 [03:00<00:52,  1.51s/it]
+
+                                                 
+{'loss': 0.7707, 'grad_norm': 0.33796223998069763, 'learning_rate': 1.267605633802817e-05, 'epoch': 4.6}
+
+
+ 77%|███████▋  | 115/150 [03:00<00:52,  1.51s/it]
+
+ 77%|███████▋  | 116/150 [03:01<00:51,  1.51s/it]
+
+ 78%|███████▊  | 117/150 [03:03<00:50,  1.52s/it]
+
+ 79%|███████▊  | 118/150 [03:04<00:48,  1.52s/it]
+
+ 79%|███████▉  | 119/150 [03:06<00:46,  1.49s/it]
+
+ 80%|████████  | 120/150 [03:07<00:45,  1.51s/it]
+
+                                                 
+{'loss': 0.7708, 'grad_norm': 0.33394232392311096, 'learning_rate': 1.0915492957746478e-05, 'epoch': 4.8}
+
+
+ 80%|████████  | 120/150 [03:07<00:45,  1.51s/it]
+
+ 81%|████████  | 121/150 [03:09<00:44,  1.53s/it]
+
+ 81%|████████▏ | 122/150 [03:10<00:43,  1.54s/it]
+
+ 82%|████████▏ | 123/150 [03:12<00:41,  1.53s/it]
+
+ 83%|████████▎ | 124/150 [03:13<00:39,  1.50s/it]
+
+ 83%|████████▎ | 125/150 [03:15<00:37,  1.51s/it]
+
+                                                 
+{'loss': 0.7812, 'grad_norm': 0.32235094904899597, 'learning_rate': 9.15492957746479e-06, 'epoch': 5.0}
+
+
+ 83%|████████▎ | 125/150 [03:15<00:37,  1.51s/it]
+
+ 84%|████████▍ | 126/150 [03:16<00:36,  1.50s/it]
+
+ 85%|████████▍ | 127/150 [03:18<00:34,  1.51s/it]
+
+ 85%|████████▌ | 128/150 [03:19<00:32,  1.49s/it]
+
+ 86%|████████▌ | 129/150 [03:21<00:31,  1.50s/it]
+
+ 87%|████████▋ | 130/150 [03:22<00:29,  1.48s/it]
+
+                                                 
+{'loss': 0.7453, 'grad_norm': 0.3531821668148041, 'learning_rate': 7.394366197183099e-06, 'epoch': 5.2}
+
+
+ 87%|████████▋ | 130/150 [03:22<00:29,  1.48s/it]
+
+ 87%|████████▋ | 131/150 [03:24<00:28,  1.51s/it]
+
+ 88%|████████▊ | 132/150 [03:25<00:27,  1.50s/it]
+
+ 89%|████████▊ | 133/150 [03:27<00:25,  1.50s/it]
+
+ 89%|████████▉ | 134/150 [03:28<00:23,  1.49s/it]
+
+ 90%|█████████ | 135/150 [03:30<00:22,  1.51s/it]
+
+                                                 
+{'loss': 0.7588, 'grad_norm': 0.3414919972419739, 'learning_rate': 5.6338028169014084e-06, 'epoch': 5.4}
+
+
+ 90%|█████████ | 135/150 [03:30<00:22,  1.51s/it]
+
+ 91%|█████████ | 136/150 [03:31<00:21,  1.51s/it]
+
+ 91%|█████████▏| 137/150 [03:33<00:19,  1.51s/it]
+
+ 92%|█████████▏| 138/150 [03:34<00:18,  1.53s/it]
+
+ 93%|█████████▎| 139/150 [03:36<00:16,  1.52s/it]
+
+ 93%|█████████▎| 140/150 [03:37<00:15,  1.54s/it]
+
+                                                 
+{'loss': 0.7859, 'grad_norm': 0.33363598585128784, 'learning_rate': 3.873239436619718e-06, 'epoch': 5.6}
+
+
+ 93%|█████████▎| 140/150 [03:37<00:15,  1.54s/it]
+
+ 94%|█████████▍| 141/150 [03:39<00:13,  1.53s/it]
+
+ 95%|█████████▍| 142/150 [03:40<00:12,  1.52s/it]
+
+ 95%|█████████▌| 143/150 [03:42<00:10,  1.52s/it]
+
+ 96%|█████████▌| 144/150 [03:43<00:09,  1.53s/it]
+
+ 97%|█████████▋| 145/150 [03:45<00:07,  1.52s/it]
+
+                                                 
+{'loss': 0.7676, 'grad_norm': 0.3417748510837555, 'learning_rate': 2.112676056338028e-06, 'epoch': 5.8}
+
+
+ 97%|█████████▋| 145/150 [03:45<00:07,  1.52s/it]
+
+ 97%|█████████▋| 146/150 [03:46<00:06,  1.51s/it]
+
+ 98%|█████████▊| 147/150 [03:48<00:04,  1.52s/it]
+
+ 99%|█████████▊| 148/150 [03:49<00:02,  1.50s/it]
+
+ 99%|█████████▉| 149/150 [03:51<00:01,  1.49s/it]
+
+100%|██████████| 150/150 [03:53<00:00,  1.53s/it]
+
+                                                 
+{'loss': 0.7767, 'grad_norm': 0.3764980137348175, 'learning_rate': 3.5211267605633803e-07, 'epoch': 6.0}
+
+
+100%|██████████| 150/150 [03:53<00:00,  1.53s/it]
+
+                                                 
+{'train_runtime': 233.6201, 'train_samples_per_second': 5.137, 'train_steps_per_second': 0.642, 'train_loss': 1.1637831672032675, 'epoch': 6.0}
+
+
+100%|██████████| 150/150 [03:53<00:00,  1.53s/it]
+100%|██████████| 150/150 [03:53<00:00,  1.56s/it]
+  SFT done in 3.9 min
+
+══ Pre-GRPO hold-out eval (SFT-only) ══
+
+  [diagnostic] seed=100 raw completion (first 500 chars):
+  </tool_call>
+  The news signals a severe supply chain fragmentation and transition shock. 1st-order: Tech faces immediate headwinds from export controls and rare-earth shortages, while OIL benefits from sector rotation. 2nd-order: GREEN faces supply chain constraints, but the transition shock provides a tailwind. 3rd-order: BONDS and REAL_ESTATE are safe havens. 
+  Given the 12-quarter lockup, I must avoid OIL due to the 25 kg carbon cap. A 0.25 allocation to OIL would emit 2.5 * 0.25 * 12 = 7.5 kg,
+  [parse_action result]: metadata={} weights=[0.35, 0.0, 0.45, 0.15, 0.05] infra_commit=0.15 carbon_offset_buy=0.0 put_hedge=0.0 tech_bet='fragmentation'
+
+── Hold-out eval (5/5 valid) ──
+  mean regret: +0.0340
+  beat baseline: 3/5
+Found HuggingFace hub cache directory: /tmp/CarbonAlpha/hf_cache/hub
+Checking cache directory for required files...
+
+
+Unsloth: Copying 2 files from cache to `/tmp/CarbonAlpha/checkpoints/final_merged`:   0%|          | 0/2 [00:00<?, ?it/s]
+
+Unsloth: Copying 2 files from cache to `/tmp/CarbonAlpha/checkpoints/final_merged`: 100%|██████████| 2/2 [00:01<00:00,  1.40it/s]
+Unsloth: Copying 2 files from cache to `/tmp/CarbonAlpha/checkpoints/final_merged`: 100%|██████████| 2/2 [00:01<00:00,  1.40it/s]
+Successfully copied all 2 files from cache to `/tmp/CarbonAlpha/checkpoints/final_merged`
+Checking cache directory for required files...
+Cache check failed: tokenizer.model not found in local cache.
+Not all required files found in cache. Will proceed with downloading.
+
+
+Unsloth: Preparing safetensor model files:   0%|          | 0/2 [00:00<?, ?it/s]
+Unsloth: Preparing safetensor model files: 100%|██████████| 2/2 [00:00<00:00, 57065.36it/s]
+
+
+Unsloth: Merging weights into 16bit:   0%|          | 0/2 [00:00<?, ?it/s]
+
+Unsloth: Merging weights into 16bit:  50%|█████     | 1/2 [00:30<00:30, 30.52s/it]
+
+Unsloth: Merging weights into 16bit: 100%|██████████| 2/2 [00:47<00:00, 22.45s/it]
+Unsloth: Merging weights into 16bit: 100%|██████████| 2/2 [00:47<00:00, 23.66s/it]
+Unsloth: Merge process complete. Saved to `/tmp/CarbonAlpha/checkpoints/final_merged`
+SFT-only mode. Saved LoRA adapters to /tmp/CarbonAlpha/checkpoints/final_merged