Upload folder using huggingface_hub (#2)

- 022e8b5a70b0de9db0a7f92927a371d3455bba13ce008e466c804e4ee581dbf0 (9e338ea555387ac238c0cfa7a205c09770dfe4fe)
- a4e94429c00c65cea339d451393ef893092f9ac28564bb6bd427aba8d770acec (7c7288fc3321f7bf99988830d699946e8c6fc1dc)
- 9e08feba75f3914336429bd550ea562e872811603d7e7f61abdea3e49674771d (f35f0ca4855998ff8830f562a0bafb5a6c11a1dc)

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-rl-bnb-8bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/deepseek-math-7b-rl")
     
     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/tmpo5lj0wj9",
+  "_name_or_path": "/tmp/tmpivugclnx",
   "architectures": [
     "LlamaForCausalLM"
   ],
@@ -21,7 +21,7 @@
   "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/modelsnnf0ia48",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsl5v_8gnb",
         "batch_size": 1,
         "model_name": "deepseek-ai/deepseek-math-7b-rl",
         "pruning_ratio": 0.0,