Exchange Llama2 against Llama3 in HuggingFace_accelerate example (#3108)

* Rename llama2 to llama in HF_accelerate

* Replace llama2 with llama3 in large_models/Huggingface_accelerate

* Remove code handler

* Fix lint error

examples/large_models/Huggingface_accelerate/Download_model.py

@@ -47,5 +47,6 @@ def hf_model(model_str):
     revision=args.revision,
     cache_dir=args.model_path,
     use_auth_token=True,
+    ignore_patterns=["original/*"],
 )
 print(f"Files for '{args.model_name}' is downloaded to '{snapshot_path}'")

examples/large_models/Huggingface_accelerate/llama2/Readme.md → examples/large_models/Huggingface_accelerate/llama/Readme.md

@@ -1,10 +1,10 @@
-# Loading meta-llama/Llama-2-70b-chat-hf on AWS EC2 g5.24xlarge using accelerate
+# Loading meta-llama/Meta-Llama-3-70B-Instruct on AWS EC2 g5.24xlarge using accelerate
 
-This document briefs on serving large HG models with limited resource using accelerate. This option can be activated with `low_cpu_mem_usage=True`. The model is first created on the Meta device (with empty weights) and the state dict is then loaded inside it (shard by shard in the case of a sharded checkpoint).
+This document briefs on serving large HF models with limited resource using accelerate. This option can be activated with `low_cpu_mem_usage=True`. The model is first created on the Meta device (with empty weights) and the state dict is then loaded inside it (shard by shard in the case of a sharded checkpoint). This examples uses Meta Llama-3 as an example but it works with Llama2 as well by replacing the model identifier.
 
 ### Step 1: Download model Permission
 
-Follow [this instruction](https://huggingface.co/meta-llama/Llama-2-70b-chat-hf) to get permission
+Follow [this instruction](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct) to get permission
 
 Login with a Hugging Face account
 ```
@@ -14,44 +14,44 @@ huggingface-cli login --token $HUGGINGFACE_TOKEN
 ```
 
 ```bash
-python ../Download_model.py --model_path model --model_name meta-llama/Llama-2-70b-chat-hf
+python ../Download_model.py --model_path model --model_name meta-llama/Meta-Llama-3-70B-Instruct
 ```
-Model will be saved in the following path, `model/models--meta-llama--Llama-2-70b-chat-hf`.
+Model will be saved in the following path, `model/models--meta-llama--Meta-Llama-3-70B-Instruct`.
 
 ### Step 2: Generate MAR file
 
 Add the downloaded path to " model_path:" in `model-config.yaml` and run the following.
 
 ```bash
-torch-model-archiver --model-name llama2-70b-chat --version 1.0 --handler custom_handler.py --config-file model-config.yaml -r requirements.txt --archive-format no-archive
+torch-model-archiver --model-name llama3-70b-instruct --version 1.0 --handler custom_handler.py --config-file model-config.yaml -r requirements.txt --archive-format no-archive
 ```
 
-If you are using conda, and notice issues with mpi4py, you would need to install openmpi-mpicc using the following
+If you are using conda, and notice issues with mpi4py, you can install it with
 
 ```
-conda install -c conda-forge openmpi-mpicc
+conda install mpi4py
 ```
 
 ### Step 3: Add the mar file to model store
 
 ```bash
 mkdir model_store
-mv llama2-70b-chat model_store
-mv model model_store/llama2-70b-chat
+mv llama3-70b-instruct model_store
+mv model model_store/llama3-70b-instruct
 ```
 
 ### Step 3: Start torchserve
 
 Update config.properties and start torchserve
 
 ```bash
-torchserve --start --ncs --ts-config config.properties --model-store model_store --models llama2-70b-chat
+torchserve --start --ncs --ts-config config.properties --model-store model_store --models llama3-70b-instruct
 ```
 
 ### Step 4: Run inference
 
 ```bash
-curl -v "http://localhost:8080/predictions/llama2-70b-chat" -T sample_text.txt
+curl -v "http://localhost:8080/predictions/llama3-70b-instruct" -T sample_text.txt
 ```
 
 results in the following output

examples/large_models/Huggingface_accelerate/llama2/custom_handler_code.py → examples/large_models/Huggingface_accelerate/llama/custom_handler.py

@@ -1,9 +1,10 @@
 import logging
 from abc import ABC
+from typing import Dict
 
 import torch
 import transformers
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 from ts.context import Context
 from ts.torch_handler.base_handler import BaseHandler
@@ -39,26 +40,30 @@ def initialize(self, ctx: Context):
         seed = int(ctx.model_yaml_config["handler"]["manual_seed"])
         torch.manual_seed(seed)
 
-        logger.info("Model %s loading tokenizer", ctx.model_name)
+        self.tokenizer = AutoTokenizer.from_pretrained(model_path)
+        self.tokenizer.pad_token = self.tokenizer.eos_token
+        self.tokenizer.padding_side = "left"
+        logger.info("Model %s loaded tokenizer successfully", ctx.model_name)
+
+        if self.tokenizer.vocab_size >= 128000:
+            quant_config = BitsAndBytesConfig(
+                load_in_4bit=True,
+                bnb_4bit_use_double_quant=True,
+                bnb_4bit_quant_type="nf4",
+                bnb_4bit_compute_dtype=torch.bfloat16,
+            )
+        else:
+            quant_config = BitsAndBytesConfig(load_in_8bit=True)
+
         self.model = AutoModelForCausalLM.from_pretrained(
             model_path,
             device_map="balanced",
             low_cpu_mem_usage=True,
             torch_dtype=torch.float16,
-            load_in_8bit=True,
+            quantization_config=quant_config,
             trust_remote_code=True,
         )
-        if ctx.model_yaml_config["handler"]["fast_kernels"]:
-            from optimum.bettertransformer import BetterTransformer
-
-            try:
-                self.model = BetterTransformer.transform(self.model)
-            except RuntimeError as error:
-                logger.warning(
-                    "HuggingFace Optimum is not supporting this model,for the list of supported models, please refer to this doc,https://huggingface.co/docs/optimum/bettertransformer/overview"
-                )
-        self.tokenizer = AutoTokenizer.from_pretrained(model_path)
-
+        self.device = next(iter(self.model.parameters())).device
         logger.info("Model %s loaded successfully", ctx.model_name)
         self.initialized = True
 
@@ -72,38 +77,31 @@ def preprocess(self, requests):
             tuple: A tuple with two tensors: the batch of input ids and the batch of
                 attention masks.
         """
-        input_texts = [data.get("data") or data.get("body") for data in requests]
-        input_ids_batch, attention_mask_batch = [], []
-        for input_text in input_texts:
-            input_ids, attention_mask = self.encode_input_text(input_text)
-            input_ids_batch.append(input_ids)
-            attention_mask_batch.append(attention_mask)
-        input_ids_batch = torch.cat(input_ids_batch, dim=0).to(self.model.device)
-        attention_mask_batch = torch.cat(attention_mask_batch, dim=0).to(self.device)
-        return input_ids_batch, attention_mask_batch
-
-    def encode_input_text(self, input_text):
+        input_texts = [self.preprocess_requests(r) for r in requests]
+
+        logger.info("Received texts: '%s'", input_texts)
+        inputs = self.tokenizer(
+            input_texts,
+            max_length=self.max_length,
+            padding=True,
+            add_special_tokens=True,
+            return_tensors="pt",
+            truncation=True,
+        ).to(self.device)
+        return inputs
+
+    def preprocess_requests(self, request: Dict):
         """
-        Encodes a single input text using the tokenizer.
+        Preprocess request
         Args:
-            input_text (str): The input text to be encoded.
+            request (Dict): Request to be decoded.
         Returns:
-            tuple: A tuple with two tensors: the encoded input ids and the attention mask.
+            str: Decoded input text
         """
+        input_text = request.get("data") or request.get("body")
         if isinstance(input_text, (bytes, bytearray)):
             input_text = input_text.decode("utf-8")
-        logger.info("Received text: '%s'", input_text)
-        inputs = self.tokenizer.encode_plus(
-            input_text,
-            max_length=self.max_length,
-            padding=False,
-            add_special_tokens=True,
-            return_tensors="pt",
-            truncation=True,
-        )
-        input_ids = inputs["input_ids"]
-        attention_mask = inputs["attention_mask"]
-        return input_ids, attention_mask
+        return input_text
 
     def inference(self, input_batch):
         """
@@ -115,11 +113,8 @@ def inference(self, input_batch):
         Returns:
             list: A list of strings with the predicted values for each input text in the batch.
         """
-        input_ids_batch, attention_mask_batch = input_batch
-        input_ids_batch = input_ids_batch.to(self.device)
         outputs = self.model.generate(
-            input_ids_batch,
-            attention_mask=attention_mask_batch,
+            **input_batch,
             max_length=self.max_new_tokens,
         )
 

examples/large_models/Huggingface_accelerate/llama2/model-config.yaml → examples/large_models/Huggingface_accelerate/llama/model-config.yaml

@@ -6,9 +6,8 @@ responseTimeout: 1200
 deviceType: "gpu"
 
 handler:
-    model_name: "meta-llama/Llama-2-70b-chat-hf"
-    model_path: "model/models--meta-llama--Llama-2-70b-chat-hf/snapshots/9ff8b00464fc439a64bb374769dec3dd627be1c2"
+    model_name: "meta-llama/Meta-Llama-3-70B-Instruct"
+    model_path: "model/models--meta-llama--Meta-Llama-3-70B-Instruct/snapshots/5fcb2901844dde3111159f24205b71c25900ffbd"
     max_length: 50
     max_new_tokens: 50
     manual_seed: 40
-    fast_kernels: True

examples/large_models/Huggingface_accelerate/llama/sample_text.txt

@@ -0,0 +1 @@
+what is the recipe of mayonnaise?

ts_scripts/spellcheck_conf/wordlist.txt

@@ -1224,3 +1224,4 @@ Fickling
 TorchServer
 VirusTotal
 untrusted
+mpi