Update model config to show correct dtype for non-quantized tensors (float16) and model name

Accidentally kept the config file from a previous version, doesn't affect loading. I want to try to do an error analysis to see whether float16 or bfloat16 makes more sense here... bfloat16 might be better.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "models/allenai_Molmo-72B-0924",
+  "_name_or_path": "SeanScripts/Molmo-72B-0924-nf4",
   "architectures": [
     "MolmoForCausalLM"
   ],
@@ -25,10 +25,10 @@
   "quantization_config": {
     "_load_in_4bit": true,
     "_load_in_8bit": false,
-    "bnb_4bit_compute_dtype": "bfloat16",
+    "bnb_4bit_compute_dtype": "float16",
     "bnb_4bit_quant_storage": "uint8",
     "bnb_4bit_quant_type": "nf4",
-    "bnb_4bit_use_double_quant": false,
+    "bnb_4bit_use_double_quant": true,
     "llm_int8_enable_fp32_cpu_offload": true,
     "llm_int8_has_fp16_weight": false,
     "llm_int8_skip_modules": ["model.wte", "model.ln_f", "model.transformer.ff_out", "model.vision_backbone"],
@@ -39,7 +39,7 @@
   },
   "rope_theta": 1000000.0,
   "tie_word_embeddings": false,
-  "torch_dtype": "bfloat16",
+  "torch_dtype": "float16",
   "transformers_version": "4.45.1",
   "use_cache": true,
   "use_position_ids": true,