Upload folder using huggingface_hub (#3)

- 1948c75b3161080d785a6eaf4fc7dd82e2e3271d10e7eb206a59be0d5025447e (8e5e6613eaf08d3ca30ec64af71aff6906d972a6)
- 237b6a912be638715b26d20906c6660b4d1f6e1fd8d0afee2d0dd31f329547e1 (e7cc6ea3c2927d2177680b5b6689bc837d4e325c)
- ad98e61d63e557db55cfc0a5e57b8b8b267d612c06def6c3ec9a98a023c8f3df (c728f73cd869494c174fdfab0105a096e03024c3)

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/scb10x-typhoon-7b-bnb-8bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("scb10x/typhoon-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/tmp25ciu4_r",
+  "_name_or_path": "/tmp/tmp68pbx112",
   "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/modelse5lr6oud",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/models83us4xvk",
         "batch_size": 1,
         "model_name": "scb10x/typhoon-7b",
         "pruning_ratio": 0.0,