Training in progress, step 16000

config.json

@@ -119,7 +119,7 @@
       "type": "audio"
     }
   },
-  "do_sample": false,
+  "do_sample": true,
   "downsample_rate": 5,
   "dtype": "bfloat16",
   "enable_thinking": true,
@@ -252,7 +252,7 @@
     "up_proj",
     "down_proj"
   ],
-  "max_new_tokens": 1024,
+  "max_new_tokens": 128,
   "min_new_tokens": 0,
   "model_dtype": "bfloat16",
   "model_type": "asr_model",
@@ -275,10 +275,10 @@
   "qformer_num_heads": 16,
   "qformer_num_layers": 2,
   "qformer_window_size": 15,
-  "repetition_penalty": 1.0,
+  "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": [
@@ -399,4 +399,4 @@
   "use_lora": false,
   "use_specaugment": true,
   "vocab_size": 128257
-}
+}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e8f40cce2270ed5e01a0055f559d9010d5043000a3c056b90acd1af8b96ba92
+oid sha256:8ac436124a77178225ce77ad718d45ef2560fdbd6090403fdc4a19794e0faa7b
 size 14682440