Training in progress, step 1000

config.json

@@ -159,7 +159,7 @@
   "inference_warmup_tokens": 10,
   "label_smoothing": 0.0,
   "length_penalty": 1.0,
-  "llm_dim": 2048,
+  "llm_dim": 1024,
   "mask_feature_length": 27,
   "mask_feature_min_masks": 2,
   "mask_feature_prob": 0.0,
@@ -179,8 +179,8 @@
   "projector_hidden_dim": null,
   "projector_init_std": 0.02,
   "projector_num_layers": 2,
-  "projector_pool_stride": 6,
-  "projector_type": "mosa",
+  "projector_pool_stride": 4,
+  "projector_type": "linear",
   "qformer_hidden_size": null,
   "qformer_intermediate_size": null,
   "qformer_num_heads": 16,
@@ -190,7 +190,7 @@
   "router_aux_loss_coef": 0.01,
   "system_prompt": "/no_think /system_override",
   "text_config": {
-    "_name_or_path": "Qwen/Qwen3-1.7B",
+    "_name_or_path": "Qwen/Qwen3-0.6B",
     "architectures": [
       "Qwen3ForCausalLM"
     ],
@@ -200,9 +200,9 @@
     "eos_token_id": 151645,
     "head_dim": 128,
     "hidden_act": "silu",
-    "hidden_size": 2048,
+    "hidden_size": 1024,
     "initializer_range": 0.02,
-    "intermediate_size": 6144,
+    "intermediate_size": 3072,
     "layer_types": [
       "full_attention",
       "full_attention",
@@ -251,7 +251,7 @@
     "use_sliding_window": false,
     "vocab_size": 151670
   },
-  "text_model_id": "Qwen/Qwen3-1.7B",
+  "text_model_id": "Qwen/Qwen3-0.6B",
   "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:fd8ccd85735b88efd3c380c00b69fcea27fb1c9f6ef9c48245fd066adab9bd68
-size 320134160
+oid sha256:a82f0e82a684298cc550fefb3a4968fa974cf79a2568d441bcf3e4b350a76f23
+size 2623704

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3345ff693dedd9814381c2dd87209e469396be74d8912ab3f03d6661ba4469b
-size 5201
+oid sha256:1207623a291f4a86f41326955931e7f95611cf08e5ed2ba9c8eeebca5dc00b18
+size 5265