Upload folder using huggingface_hub (#3)

- 149c45d36c3ebd6ed9f1c565ecc494a1ba89b5a02df795d0d07582bd6f0bc7c1 (0c912c6b30fd7bb74da0c482b525fff5644e95a7)
- 571a2ab9116c4f5c2779c8d23f196a022968fee2dbe6978e1428ec24adf38a1f (0e6925bb630c11400a2ce55d08574abb7b567921)
- 98c95c7285396394ff18b0db7baac8b5ed50fd30d472029450bd9e1a2d561ee3 (d1210a09060db05849183bedd22835884a48eb12)

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/deepseek-ai-deepseek-math-7b-instruct-bnb-8bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/deepseek-math-7b-instruct")
     
     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/tmpkeiv43ib",
+  "_name_or_path": "/tmp/tmp3nm9xn0k",
   "architectures": [
     "LlamaForCausalLM"
   ],
@@ -20,7 +20,7 @@
   "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/models4g9lz34c",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsopkm_1s5",
         "batch_size": 1,
         "model_name": "deepseek-ai/deepseek-math-7b-instruct",
         "pruning_ratio": 0.0,