GRPO checkpoint step 2 @ 2026-04-26T11:06:07Z

checkpoint-2/adapter_config.json

@@ -34,12 +34,12 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "k_proj",
-    "down_proj",
-    "gate_proj",
     "up_proj",
-    "q_proj",
     "v_proj",
+    "gate_proj",
+    "q_proj",
+    "k_proj",
+    "down_proj",
     "o_proj"
   ],
   "target_parameters": null,

checkpoint-2/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1386dbf221f290d2589933056a1ffdd6fd4bad8e8f50bd3a549b19b4304de0d
+oid sha256:43a1e6226d2b1888c323df3932874257b166314620f525ff452805ec4083258b
 size 137716248

checkpoint-2/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f0ce576ee5b94c0eac3f591c514eb71268a898a47cfc4f47cf5be79597b13ea
+oid sha256:a3e22a30ab65d8a2edee7d7fb77505b548cc5bc1042a589cb5ba287abbb05efd
 size 140572180

checkpoint-2/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:638a667570bdda9e9b7837029c18bc16f96f1d378af8113e22f7fabfb1df3aac
+oid sha256:f0f7c3ddec04837ad0d9008c37744f11293b90a8d817ddebe66c8fb50a5cfbfa
 size 14244

checkpoint-2/trainer_state.json

@@ -16,25 +16,25 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.875,
-      "completions/max_length": 768.0,
-      "completions/max_terminated_length": 423.0,
-      "completions/mean_length": 724.875,
-      "completions/mean_terminated_length": 423.0,
-      "completions/min_length": 423.0,
-      "completions/min_terminated_length": 423.0,
-      "entropy": 0.738787330687046,
+      "completions/max_length": 1536.0,
+      "completions/max_terminated_length": 859.0,
+      "completions/mean_length": 1451.375,
+      "completions/mean_terminated_length": 859.0,
+      "completions/min_length": 859.0,
+      "completions/min_terminated_length": 859.0,
+      "entropy": 0.8951552212238312,
       "epoch": 0.5,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.75,
+      "grad_norm": 0.0283203125,
       "learning_rate": 0.0,
       "loss": 0.0,
-      "num_tokens": 13245.0,
-      "reward": -0.7137750387191772,
-      "reward_std": 1.8349467515945435,
-      "rewards/reward_fn/mean": -0.7137750387191772,
-      "rewards/reward_fn/std": 1.8349467515945435,
+      "num_tokens": 19065.0,
+      "reward": -0.560462474822998,
+      "reward_std": 1.7301521301269531,
+      "rewards/reward_fn/mean": -0.560462474822998,
+      "rewards/reward_fn/std": 1.7301521301269531,
       "step": 1,
-      "step_time": 162.93437966145575
+      "step_time": 323.24305043928325
     },
     {
       "clip_ratio/high_max": 0.0,
@@ -43,30 +43,30 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.75,
-      "completions/max_length": 768.0,
-      "completions/max_terminated_length": 681.0,
-      "completions/mean_length": 703.25,
-      "completions/mean_terminated_length": 509.0,
-      "completions/min_length": 337.0,
-      "completions/min_terminated_length": 337.0,
-      "entropy": 0.9916174560785294,
+      "completions/max_length": 1536.0,
+      "completions/max_terminated_length": 1217.0,
+      "completions/mean_length": 1361.5,
+      "completions/mean_terminated_length": 838.0,
+      "completions/min_length": 459.0,
+      "completions/min_terminated_length": 459.0,
+      "entropy": 0.6386842429637909,
       "epoch": 1.0,
       "frac_reward_zero_std": 0.25,
-      "grad_norm": 0.0830078125,
+      "grad_norm": 0.051025390625,
       "learning_rate": 1e-05,
-      "loss": 0.0432,
-      "num_tokens": 26287.0,
-      "reward": -0.8818060159683228,
-      "reward_std": 1.7631477117538452,
-      "rewards/reward_fn/mean": -0.8818060159683228,
-      "rewards/reward_fn/std": 1.7631477117538452,
+      "loss": -0.0492,
+      "num_tokens": 37381.0,
+      "reward": -0.9206835031509399,
+      "reward_std": 2.0997297763824463,
+      "rewards/reward_fn/mean": -0.9206835031509399,
+      "rewards/reward_fn/std": 2.0997297763824463,
       "step": 2,
-      "step_time": 160.85512119717896
+      "step_time": 322.2724595889449
     }
   ],
   "logging_steps": 1,
   "max_steps": 2,
-  "num_input_tokens_seen": 26287,
+  "num_input_tokens_seen": 37381,
   "num_train_epochs": 1,
   "save_steps": 50,
   "stateful_callbacks": {

checkpoint-2/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fa7d5fa0cfa70849cd7f8c695cef7dd244b80503646d95af67c95739193da16
+oid sha256:2e4d13ee818464bf5a003209fc44d2a92f16d9f3bab06dd55abd6baeef04a646
 size 7352