Upload folder using huggingface_hub (#2)

- 7db9a1c7a567d35549c88208b83bf27d3a5bfaa72c32fe50262b80a241d56f34 (1666599db73491e8093a82c2877f4f326f3ab70e)
- 6e3530b2e2b486af6fdf7c232705bbbd707cb1047468e9708ba9fa39e5c96680 (7251ef63287ac6361bf7a453d09e4db654bd1665)
- f568b8c431c402b9d6e902419de520f84647b3b08f0c5c3a439d3b4be93d1b4c (dec92510cd2f0e9b2c56358d7e103e31e9a21cd1)

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/tokyotech-llm-Swallow-7b-hf-bnb-8bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("tokyotech-llm/Swallow-7b-hf")
     
     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/tmpchf84xyp",
+  "_name_or_path": "/tmp/tmp3863i5__",
   "architectures": [
     "LlamaForCausalLM"
   ],
@@ -22,7 +22,7 @@
   "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/modelsolfbmmbv",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsv58bne7z",
         "batch_size": 1,
         "model_name": "tokyotech-llm/Swallow-7b-hf",
         "pruning_ratio": 0.0,