Upload folder using huggingface_hub (#3)

- f5e171fd08386e73f497e72a3f41eccf4a918af53a27968a34c6704b18d4ac46 (a5f5ec31a7773d53cc9bbf0e4f46c12547bb0e03)
- e39a66ad0ff497f1b63e661124183d36da69de07c611596577067e3f657f5187 (0fca917b8d6455d5ff5391504999b485b5cb32a0)

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-4bit-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/tmpgn90407d",
+  "_name_or_path": "/tmp/tmp1xuobkbx",
   "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": [

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb3a739a2fb79fa5aa0f0a261733f9ad3f10d363336fb2f113d114fe20e93ffd
-size 4049149733
+oid sha256:c0b3a4a079d18c5533077149fb531d1591caceee23416500af13d6f779cdb60c
+size 4350829736

smash_config.json

@@ -8,7 +8,7 @@
         "compilers": "None",
         "task": "text_text_generation",
         "device": "cuda",
-        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsmrijn4_n",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/models5q0eih8b",
         "batch_size": 1,
         "model_name": "tokyotech-llm/Swallow-7b-hf",
         "pruning_ratio": 0.0,