Fix examples

example.py

@@ -9,12 +9,10 @@
 examples = ["auto_fp8 is an easy-to-use model quantization library"]
 examples = tokenizer(examples, return_tensors="pt").to("cuda")
 
-ignore_patterns = ["re:.*gate"]
-
 quantize_config = BaseQuantizeConfig(
     quant_method="fp8",
     activation_scheme="dynamic",  # or "static"
-    ignore_patterns=ignore_patterns,
+    ignore_patterns=["re:.*lm_head"],
 )
 
 model = AutoFP8ForCausalLM.from_pretrained(

example_dataset.py

@@ -7,24 +7,11 @@
 quantized_model_dir = "Meta-Llama-3-8B-Instruct-FP8"
 
 tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True)
+ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(512)
+examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
+examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")
 
-DATASET_ID = "mgoin/ultrachat_2k"
-DATASET_SPLIT = "train_sft"
-ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
-ds = ds.map(
-    lambda batch: {
-        "text": tokenizer.apply_chat_template(batch["messages"], tokenize=False)
-    }
-)
-examples = [sample["text"] for sample in ds]
-tokenizer.pad_token = tokenizer.eos_token
-examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to(
-    "cuda"
-)
-
-quantize_config = BaseQuantizeConfig(
-    quant_method="fp8", activation_scheme="static"
-)  # or "static"
+quantize_config = BaseQuantizeConfig(quant_method="fp8", activation_scheme="static")
 
 model = AutoFP8ForCausalLM.from_pretrained(
     pretrained_model_dir, quantize_config=quantize_config

examples/example_mixtral.py

@@ -0,0 +1,24 @@
+from datasets import load_dataset
+from transformers import AutoTokenizer
+
+from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
+
+pretrained_model_dir = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+quantized_model_dir = "Mixtral-8x7B-Instruct-v0.1-FP8"
+
+tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True)
+ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(10)
+examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
+examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")
+
+quantize_config = BaseQuantizeConfig(
+    quant_method="fp8",
+    activation_scheme="static",
+    ignore_patterns=["re:.*lm_head", "re:.*gate"],
+)
+
+model = AutoFP8ForCausalLM.from_pretrained(
+    pretrained_model_dir, quantize_config=quantize_config
+)
+model.quantize(examples)
+model.save_quantized(quantized_model_dir)