Upload folder using huggingface_hub (#2)

- fba1ce891af4f27534bbc1bc01a3aae6dbfb5d92aa06043569818aef40bf7b55 (ee19df31f899045e96eb340f417778a07add4a75)
- 97216e8e5e8c9507e4ef8ebc0c603e6cf6c38080e86b632bf9b623f99d9e1c99 (4f440c06fdb7484b8ceaf2fe5d28ef40acccf38d)
- 0409c93fcb8d3d6e8897440850c8cc9549152301c0eb94f8b603749803b76c73 (6307c6ca4cd10a7690d4560ab0538179305574f1)

README.md

@@ -34,7 +34,7 @@ tags:
 
 ## Results
 
-Detailed efficiency metrics coming soon!
+![image info](./plots.png)
 
 **Frequently Asked Questions**
 - ***How does the compression work?*** The model is compressed with llm-int8.
@@ -61,7 +61,7 @@ You can run the smashed model with these steps:
    from transformers import AutoModelForCausalLM, AutoTokenizer
 
     model = AutoModelForCausalLM.from_pretrained("PrunaAI/meta-math-MetaMath-Mistral-7B-bnb-8bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("meta-math/MetaMath-Mistral-7B")
     
     input_ids = tokenizer("What is the color of prunes?,", return_tensors='pt').to(model.device)["input_ids"]

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/tmp/tmpu_urg1yh",
+  "_name_or_path": "/tmp/tmp8mmuzs6_",
   "architectures": [
     "MistralForCausalLM"
   ],
@@ -18,7 +18,7 @@
   "quantization_config": {
     "bnb_4bit_compute_dtype": "bfloat16",
     "bnb_4bit_quant_type": "fp4",
-    "bnb_4bit_use_double_quant": true,
+    "bnb_4bit_use_double_quant": false,
     "llm_int8_enable_fp32_cpu_offload": false,
     "llm_int8_has_fp16_weight": false,
     "llm_int8_skip_modules": [

smash_config.json

@@ -8,7 +8,7 @@
         "compilers": "None",
         "task": "text_text_generation",
         "device": "cuda",
-        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsg7wd83pj",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelswq_f029c",
         "batch_size": 1,
         "model_name": "meta-math/MetaMath-Mistral-7B",
         "pruning_ratio": 0.0,