3dc5eae27faca490b6ca87a0dd7a6bbcad85ccec4f07c3d94e19bff02116206b

README.md

@@ -61,7 +61,7 @@ You can run the smashed model with these steps:
    from transformers import AutoModelForCausalLM, AutoTokenizer
 
     model = AutoModelForCausalLM.from_pretrained("PrunaAI/42dot-42dot_LLM-PLM-1.3B-bnb-4bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("42dot/42dot_LLM-PLM-1.3B")
     
     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/tmpahl9ahxq",
+  "_name_or_path": "/tmp/tmpor5u0g_u",
   "architectures": [
     "LlamaForCausalLM"
   ],

smash_config.json

@@ -2,18 +2,18 @@
     "api_key": null,
     "verify_url": "http://johnrachwan.pythonanywhere.com",
     "smash_config": {
-        "pruners": "[]",
-        "factorizers": "[]",
+        "pruners": "None",
+        "factorizers": "None",
         "quantizers": "['llm-int8']",
-        "compilers": "[]",
+        "compilers": "None",
         "task": "text_text_generation",
         "device": "cuda",
-        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsq_fbgga3",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelsz_jn2cg7",
         "batch_size": 1,
         "model_name": "42dot/42dot_LLM-PLM-1.3B",
         "pruning_ratio": 0.0,
         "n_quantization_bits": 4,
-        "output_deviation": 0.01,
+        "output_deviation": 0.005,
         "max_batch_size": 1,
         "qtype_weight": "torch.qint8",
         "qtype_activation": "torch.quint8",