Upload folder using huggingface_hub (#2)

- 15dc8cc9c6c1633a208d9ec63151da1b8bcdfcea1f4fbc3d5260cccc2b640a3f (6b495208d483824b4aec4614c207744e9fb915e5)
- 10e812014f042e4fc129d424798b4772d01e902196fe40f1e8bf4da865c808f8 (ffaf2f51d8bc7f6991dd3fb875ac9a3e54ad5d67)

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/MediaTek-Research-Breeze-7B-Instruct-v1_0-bnb-4bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("MediaTek-Research/Breeze-7B-Instruct-v1_0")
     
     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/tmpzxut_5f8",
+  "_name_or_path": "/tmp/tmpxorsimq6",
   "architectures": [
     "MistralForCausalLM"
   ],
@@ -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": [

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3884c7a61b9bd51868706be05742e26447ac8f2628d7d0d857dee47720f10584
-size 4616421477
+oid sha256:a0674c6b8de43bc82f81358f3b2ab59ab64c62fbd7331693d141a8e8638e5b24
+size 4941571200

smash_config.json

@@ -8,7 +8,7 @@
         "compilers": "None",
         "task": "text_text_generation",
         "device": "cuda",
-        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelssq3_ni6r",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsa_y4ult5",
         "batch_size": 1,
         "model_name": "MediaTek-Research/Breeze-7B-Instruct-v1_0",
         "pruning_ratio": 0.0,