step 10000

config.json

@@ -87,6 +87,7 @@
     "vocab_size": 151936
   },
   "base_model_name_or_path": "Qwen/Qwen3-8B",
+  "convergence_epsilon": 0.01,
   "decoder_layer_indices": [
     35
   ],
@@ -100,8 +101,10 @@
     6,
     7
   ],
+  "halting_strategy": "threshold",
   "hidden_size": 4096,
   "model_type": "lds",
+  "q_eval_interval": 1,
   "q_threshold": 0.9,
   "reasoning_layer_indices": [
     8,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac289ef85ce448872b34d79b622ed497a03cf8df6d6f06bdaacd9110f3be7bf2
-size 16389989668
+oid sha256:2713b65854a2625f702b4f5cdb74199cfca4cea544e1c6390bfa438f5e0b5f1b
+size 16385754482

tokenizer_config.json

@@ -20,7 +20,7 @@
     "<|image_pad|>",
     "<|video_pad|>"
   ],
-  "is_local": false,
+  "is_local": true,
   "model_max_length": 131072,
   "pad_token": "<|endoftext|>",
   "split_special_tokens": false,