Training in progress, step 500

config.json

@@ -160,7 +160,7 @@
   "inference_warmup_tokens": 10,
   "label_smoothing": 0.0,
   "length_penalty": 1.0,
-  "llm_dim": 1024,
+  "llm_dim": 2048,
   "max_new_tokens": 96,
   "model_dtype": "bfloat16",
   "model_type": "asr_model",
@@ -185,7 +185,7 @@
   "router_aux_loss_coef": 0.01,
   "system_prompt": "/no_think /system_override",
   "text_config": {
-    "_name_or_path": "Qwen/Qwen3-0.6B",
+    "_name_or_path": "Qwen/Qwen3-1.7B",
     "architectures": [
       "Qwen3ForCausalLM"
     ],
@@ -195,9 +195,9 @@
     "eos_token_id": 151645,
     "head_dim": 128,
     "hidden_act": "silu",
-    "hidden_size": 1024,
+    "hidden_size": 2048,
     "initializer_range": 0.02,
-    "intermediate_size": 3072,
+    "intermediate_size": 6144,
     "layer_types": [
       "full_attention",
       "full_attention",
@@ -246,7 +246,7 @@
     "use_sliding_window": false,
     "vocab_size": 151670
   },
-  "text_model_id": "Qwen/Qwen3-0.6B",
+  "text_model_id": "Qwen/Qwen3-1.7B",
   "transformers_version": "5.0.0.dev0",
   "use_cache": false,
   "use_specaugment": true,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4f1083de8fd34af265b1d0a74a11b9b907782012a3c26f2eeb1bfe1dce6451a
-size 12583152
+oid sha256:75181d026149248d64cd1870a42401228a0dffb60c08d1c6ebd934ea2d642a47
+size 58732960