Update model with corrected safetensors metadata and proper latent attention loading

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "lukeingawesome/llm2vec4cxr",
+  "_name_or_path": "microsoft/LLM2CLIP-Llama-3.2-1B-Instruct-CC-Finetuned",
   "architectures": [
     "LlamaBiModel"
   ],
@@ -8,7 +8,6 @@
   "auto_map": {
     "AutoModel": "microsoft/LLM2CLIP-Llama-3.2-1B-Instruct-CC-Finetuned--modeling_bidirectional_llama_encoder.LlamaBiModel"
   },
-  "base_model": "microsoft/LLM2CLIP-Llama-3.2-1B-Instruct-CC-Finetuned",
   "bos_token_id": 128000,
   "eos_token_id": [
     128001,
@@ -26,7 +25,6 @@
   "num_attention_heads": 32,
   "num_hidden_layers": 16,
   "num_key_value_heads": 8,
-  "pooling_mode": "latent_attention",
   "pretraining_tp": 1,
   "rms_norm_eps": 1e-05,
   "rope_scaling": {
@@ -41,8 +39,5 @@
   "torch_dtype": "bfloat16",
   "transformers_version": "4.44.2",
   "use_cache": true,
-  "vocab_size": 128256,
-  "model_description": "Fine-tuned LLM2Vec model for chest X-ray report analysis with latent attention pooling",
-  "domain": "medical",
-  "task": "text-embeddings"
-}
+  "vocab_size": 128256
+}

llm2vec_config.json

@@ -0,0 +1,6 @@
+{
+  "pooling_mode": "latent_attention",
+  "max_length": 512,
+  "doc_max_length": null,
+  "skip_instruction": false
+}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f692b8f91c3d873f7cfe1e96b40787ed73651492dd33ec877c5ff34e896ab13
-size 2524099888
+oid sha256:3de35777475189d938e3a0427896e498b22360ebbc7536065f8198e544c878cb
+size 2524099968