SFT Update: Enhanced CoT reasoning and format enforcement

config.json

@@ -5,7 +5,7 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "bos_token_id": 151643,
-  "dtype": "float16",
+  "dtype": "bfloat16",
   "eos_token_id": 151645,
   "head_dim": 128,
   "hidden_act": "silu",

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:578fb39c862d191f9e71b3efb7658f7a16b9a45d86b097905bacb12e3477ce64
-size 4902257592
+oid sha256:478a9001a2b2b77003265a861e4266d583dc1aaaa8f445dd0d993cd77ec8efc3
+size 4902257696

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33d60d555c9b1f18b5d9b47ec583a94078154a39f03cb6209cb2a34760b3c4e4
-size 4915960224
+oid sha256:e757ce93cbce4e392d88586ee066d7c212447713329ba2fb8e86179f7383ffcb
+size 4915960368

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d19c0040d2cea5d530f84541ac89a42856abddad89dfcf963c8773c7bb0e883
-size 4983068360
+oid sha256:2cd5e77ad9b83a9f6150b7c3290cb064f452c623f1bfbc10eae7b5d3269e0f52
+size 4983068496

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b99c222ff6ad2e3cc67604ba9f224cdffef8aceae0d1e63276ad4e739bd4bf6b
-size 1580230248
+oid sha256:2b1eaba4e0d04fcc0141fae86280d66003f103dada77c4db70f29ab3fd70c9a3
+size 1580230264

training_info.json

@@ -0,0 +1,6 @@
+{
+  "version": "qwen3_sft_merged",
+  "base_model": "Kiy-K/Fyodor-Q3-8B-Instruct",
+  "checkpoint": "/content/drive/MyDrive/fyodor-checkpoints/q3-sft/best_step600/checkpoint.pt",
+  "final_loss": 0.7200766705331348
+}