Training in progress, step 19000

config.json

@@ -119,10 +119,10 @@
       "type": "audio"
     }
   },
-  "do_sample": false,
+  "do_sample": true,
   "downsample_rate": 5,
   "dtype": "bfloat16",
-  "enable_thinking": false,
+  "enable_thinking": true,
   "encoder": {
     "_name_or_path": "zai-org/GLM-ASR-Nano-2512",
     "architectures": [
@@ -252,7 +252,7 @@
     "up_proj",
     "down_proj"
   ],
-  "max_new_tokens": 64,
+  "max_new_tokens": 128,
   "min_new_tokens": 0,
   "model_dtype": "bfloat16",
   "model_type": "asr_model",
@@ -278,7 +278,7 @@
   "repetition_penalty": 1.1,
   "router_aux_loss_coef": 0.01,
   "system_prompt": "",
-  "temperature": 0.8,
+  "temperature": 1.0,
   "text_config": {
     "_name_or_path": "HuggingFaceTB/SmolLM3-3B",
     "architectures": [
@@ -392,11 +392,11 @@
   },
   "text_model_id": "HuggingFaceTB/SmolLM3-3B",
   "time_mask_length": 100,
-  "top_k": 20,
-  "top_p": 0.1,
+  "top_k": 0,
+  "top_p": 0.01,
   "transformers_version": "5.0.0",
   "use_cache": false,
   "use_lora": false,
   "use_specaugment": true,
   "vocab_size": 128257
-}
+}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b58b9529fa010895757f2c16f0efd5c5e96babfa802a0a3eaecced1d1035eb78
+oid sha256:bdc328fd173116582a1b8d9c999bb438e9b939494190e5840ae74a3974173f5c
 size 14682440