Upload folder using huggingface_hub (#2)

- 74c72a8a8bbd5e46f7f54629aab860c5703eb0cb1d7d1dcb9d644d78a61b26bb (193d4e53724773a06166d55a8a63388aff92966b)
- b968a5d4567ef2d4fa1f0cad346967106a60954ebd6f1928ecbb0358a34b8681 (280b67b50f2190cfa6679b2929e4f9df40d2478c)
- d45774a86e3fab29f0a92ef1ec7a9b42c314936d1f4c148802578b09ba46bdfc (86bccb51891dbcedc0135a280b4962452f04c8a6)

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/euclaise-Ferret_7B-bnb-8bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("euclaise/Ferret_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/tmp21uyjg2y",
+  "_name_or_path": "/tmp/tmpl_i7g7yp",
   "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/modelsm55vu15_",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsj3nk7mie",
         "batch_size": 1,
         "model_name": "euclaise/Ferret_7B",
         "pruning_ratio": 0.0,