Training in progress, step 1000

config.json

@@ -236,7 +236,6 @@
   "encoder_dim": 1280,
   "freeze_projector": false,
   "freq_mask_length": 27,
-  "inference_warmup_tokens": 10,
   "label_smoothing": 0.0,
   "length_penalty": 1.0,
   "llm_dim": 2048,
@@ -252,7 +251,7 @@
     "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 +277,7 @@
   "repetition_penalty": 1.1,
   "router_aux_loss_coef": 0.01,
   "system_prompt": "",
-  "temperature": 0.8,
+  "temperature": null,
   "text_config": {
     "_name_or_path": "HuggingFaceTB/SmolLM3-3B",
     "architectures": [
@@ -392,11 +391,11 @@
   },
   "text_model_id": "HuggingFaceTB/SmolLM3-3B",
   "time_mask_length": 100,
-  "top_k": 20,
-  "top_p": 0.1,
+  "top_k": null,
+  "top_p": null,
   "transformers_version": "5.0.0",
   "use_cache": false,
   "use_lora": false,
   "use_specaugment": true,
   "vocab_size": 128257
-}
+}

generation_config.json

@@ -1,6 +1,6 @@
 {
   "bos_token_id": 128000,
-  "do_sample": true,
+  "do_sample": false,
   "eos_token_id": [
     128012
   ],
@@ -11,9 +11,6 @@
   "num_beams": 1,
   "pad_token_id": 128004,
   "repetition_penalty": 1.1,
-  "temperature": 1.0,
-  "top_k": 0,
-  "top_p": 0.01,
   "transformers_version": "5.0.0",
   "use_cache": true
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bdc328fd173116582a1b8d9c999bb438e9b939494190e5840ae74a3974173f5c
+oid sha256:4dfa2853ff5542905d337e8aeff8cbf60143d732b14880ef5d06688db14f9b80
 size 14682440