upload-4-bit

.ipynb_checkpoints/config-checkpoint.json

@@ -0,0 +1,57 @@
+{
+  "_name_or_path": "../sample_models/ArmoRM-llama3/",
+  "architectures": [
+    "LlamaForRewardModelWithGating"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoModelForSequenceClassification": "modeling_custom.LlamaForRewardModelWithGating"
+  },
+  "bos_token_id": 128000,
+  "eos_token_id": 128001,
+  "gating_hidden_dim": 1024,
+  "gating_n_hidden": 3,
+  "gating_temperature": 10,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "id2label": {
+    "0": "LABEL_0"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "label2id": {
+    "LABEL_0": 0
+  },
+  "max_position_embeddings": 8192,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "num_objectives": 19,
+  "pad_token_id": 128256,
+  "pretraining_tp": 1,
+  "quantization_config": {
+    "_load_in_4bit": true,
+    "_load_in_8bit": false,
+    "bnb_4bit_compute_dtype": "bfloat16",
+    "bnb_4bit_quant_storage": "uint8",
+    "bnb_4bit_quant_type": "nf4",
+    "bnb_4bit_use_double_quant": true,
+    "llm_int8_enable_fp32_cpu_offload": false,
+    "llm_int8_has_fp16_weight": false,
+    "llm_int8_skip_modules": null,
+    "llm_int8_threshold": 6.0,
+    "load_in_4bit": true,
+    "load_in_8bit": false,
+    "quant_method": "bitsandbytes"
+  },
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 500000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.40.2",
+  "use_cache": false,
+  "vocab_size": 128257
+}

README.md

@@ -35,24 +35,12 @@ license: llama3
 ## Demo Code
 ```python
 import torch
-from transformers import AutoConfig, AutoModelForSequenceClassification
-from transformers import BitsAndBytesConfig
-from transformers import AutoTokenizer, pipeline
-
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
 device = "cuda"
-path = "SteveTran/ArmoRM-Llama3-8B-v0.1-4bit"
+path = "RLHFlow/ArmoRM-Llama3-8B-v0.1"
+model = AutoModelForSequenceClassification.from_pretrained(path, device_map=device, 
+                               trust_remote_code=True, torch_dtype=torch.bfloat16)
 tokenizer = AutoTokenizer.from_pretrained(path, use_fast=True)
-bnb_quantization_config = BitsAndBytesConfig(load_in_4bit=True, 
-                                                bnb_4bit_compute_dtype=torch.bfloat16,
-                                                bnb_4bit_quant_type="fp4", 
-                                                bnb_4bit_use_double_quant=True)
-model = AutoModelForSequenceClassification.from_pretrained(
-    new_weights_location, 
-    quantization_config=bnb_quantization_config,
-    device_map="auto", 
-    torch_dtype=torch.bfloat16,     
-    trust_remote_code=True,
-)
 # We load a random sample from the validation set of the HelpSteer dataset
 prompt = 'What are some synonyms for the word "beautiful"?'
 response = "Nicely, Beautifully, Handsome, Stunning, Wonderful, Gorgeous, Pretty, Stunning, Elegant"

config.json

@@ -36,7 +36,7 @@
     "_load_in_8bit": false,
     "bnb_4bit_compute_dtype": "bfloat16",
     "bnb_4bit_quant_storage": "uint8",
-    "bnb_4bit_quant_type": "fp4",
+    "bnb_4bit_quant_type": "nf4",
     "bnb_4bit_use_double_quant": true,
     "llm_int8_enable_fp32_cpu_offload": false,
     "llm_int8_has_fp16_weight": false,

model-00001-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7002cddb2ab8d0f289596cc337c78b5f4c482b0e31229f88ff4efee8762c9f13
-size 2985501395
+oid sha256:e2d30bb5db48afcb66c8663120ddf2097610c5c5494063a7bedab000991ba6e9
+size 2985566107

model-00002-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42b4ca78d243ecb3c5917ec35ca07b3256e70ddf1baa79096f37fe74b2efcfb6
-size 1669796944
+oid sha256:3e5aa6b7fbc4a6d786894f52b8304a09e6f0fe4000dd92065672bff7897be56c
+size 1669853624