Fix SQ baichuan without position_ids for torch and ipex 2.3.0 (#1597)

Signed-off-by: Wang, Chang <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
intel · Jun 11, 2024 · 14734de · 14734de
1 parent 3f492c4
commit 14734de
Show file tree

Hide file tree

Showing 7 changed files with 63 additions and 91 deletions.
diff --git a/examples/.config/pytorch_optimize.json b/examples/.config/pytorch_optimize.json
@@ -2268,32 +2268,6 @@
       }
     }
   },
-  "baichuan_7b_gen_ipex_static": {
-    "working_dir": "huggingface/pytorch/text-generation/quantization",
-    "tune": {
-      "cmd": "bash run_tuning.sh",
-      "params": {
-        "topology": "baichuan_7b",
-        "task": "generation",
-        "approach": "static",
-        "output_model": "saved_results"
-      }
-    },
-    "benchmark": {
-      "cmd": "bash run_benchmark.sh",
-      "params": {
-        "topology": "baichuan_7b",
-        "task": "generation",
-        "approach": "static",
-        "backend": "ipex",
-        "mode": "benchmark",
-        "batch_size": "112",
-        "iters": "100",
-        "int8": "false",
-        "config": "saved_results"
-      }
-    }
-  },
   "baichuan2_7b_gen_ipex_static": {
     "working_dir": "huggingface/pytorch/text-generation/quantization",
     "tune": {

diff --git a/examples/huggingface/pytorch/text-generation/quantization/requirements_sq.txt b/examples/huggingface/pytorch/text-generation/quantization/requirements_sq.txt
@@ -5,7 +5,7 @@ protobuf
 sentencepiece != 0.1.92
 --extra-index-url https://download.pytorch.org/whl/cpu
 torch==2.3.0+cpu
-transformers
+transformers==4.38.1
 intel_extension_for_pytorch==2.3.0
 optimum-intel==1.16.1
 bitsandbytes  #baichuan

diff --git a/examples/huggingface/pytorch/text-generation/quantization/run_benchmark.sh b/examples/huggingface/pytorch/text-generation/quantization/run_benchmark.sh
@@ -119,14 +119,12 @@ function run_benchmark {
     elif [ "${topology}" = "llama_7b" ]; then
         model_name_or_path="meta-llama/Llama-2-7b-chat-hf"
         script="run_generation_sq.py"
-        pip install transformers==4.35.2
     elif [ "${topology}" = "llama2_7b_gptq" ]; then
         model_name_or_path="meta-llama/Llama-2-7b-hf"
         script="run_generation_cpu_woq.py"
     elif [ "${topology}" = "llama_13b" ]; then
         model_name_or_path="meta-llama/Llama-2-13b-chat-hf"
         script="run_generation_sq.py"
-        pip install transformers==4.35.2
     elif [ "${topology}" = "dolly_v2_3b" ]; then
         model_name_or_path="/tf_dataset2/models/pytorch/dolly_v2_3b"
         script="run_generation_sq.py"
@@ -137,47 +135,32 @@ function run_benchmark {
         model_name_or_path="THUDM/chatglm3-6b"
         script="run_generation_sq.py"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.35.2
     elif [ "${topology}" = "chatglm2_6b" ]; then
         model_name_or_path="THUDM/chatglm2-6b"
         script="run_generation_sq.py"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.35.2
     elif [ "${topology}" = "chatglm_6b" ]; then
         model_name_or_path="THUDM/chatglm-6b"
         script="run_generation_sq.py"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.33
     elif [ "${topology}" = "falcon_7b" ]; then
         model_name_or_path="tiiuae/falcon-7b-instruct"
         script="run_generation_sq.py"
-        pip install transformers==4.33
-    elif [ "${topology}" = "baichuan_7b" ]; then
-        model_name_or_path="baichuan-inc/Baichuan-7B"
-        extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.33
-        script="run_generation_sq.py"
     elif [ "${topology}" = "baichuan_13b" ]; then
-        model_name_or_path="baichuan-inc/Baichuan-13B-Base"
+        model_name_or_path="baichuan-inc/Baichuan-13B-Chat"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        extra_cmd=$extra_cmd" --_commit_hash 14d5b0e204542744900f6fb52422c6d633bdcb00"
-        pip install transformers==4.33
         script="run_generation_sq.py"
     elif [ "${topology}" = "baichuan2_7b" ]; then
-        model_name_or_path="baichuan-inc/Baichuan2-7B-Base"
+        model_name_or_path="baichuan-inc/Baichuan2-7B-Chat"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.33
         script="run_generation_sq.py"
     elif [ "${topology}" = "baichuan2_13b" ]; then
-        model_name_or_path="baichuan-inc/Baichuan2-13B-Base"
+        model_name_or_path="baichuan-inc/Baichuan2-13B-Chat"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.35.2
         script="run_generation_sq.py"
     elif [ "${topology}" = "qwen_7b" ]; then
-        model_name_or_path="Qwen/Qwen-7B"
+        model_name_or_path="Qwen/Qwen-7B-Chat"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        extra_cmd=$extra_cmd" --_commit_hash f7bc352f27bb1c02ee371a4576942a7d96c8bb97"
-	    pip install transformers==4.35.2
         script="run_generation_sq.py"
     elif [ "${topology}" = "mistral_7b" ]; then
         model_name_or_path="Intel/neural-chat-7b-v3"

diff --git a/examples/huggingface/pytorch/text-generation/quantization/run_tuning.sh b/examples/huggingface/pytorch/text-generation/quantization/run_tuning.sh
@@ -133,15 +133,13 @@ function run_tuning {
         model_name_or_path="/tf_dataset2/models/nlp_toolkit/llama-2-7b-chat/Llama-2-7b-chat-hf"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
-        pip install transformers==4.35.2
         script="run_generation_sq.py"
     elif [ "${topology}" = "llama_13b" ]; then
         alpha=0.8
         model_name_or_path="meta-llama/Llama-2-13b-chat-hf"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         script="run_generation_sq.py"
-        pip install transformers==4.35.2
     elif [ "${topology}" = "dolly_v2_3b" ]; then
         alpha=0.6
         model_name_or_path="/tf_dataset2/models/pytorch/dolly_v2_3b"
@@ -161,72 +159,54 @@ function run_tuning {
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
         script="run_generation_sq.py"
-        pip install transformers==4.35.2
     elif [ "${topology}" = "chatglm2_6b" ]; then
         alpha=0.75
         model_name_or_path="THUDM/chatglm2-6b"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
         script="run_generation_sq.py"
-        pip install transformers==4.35.2
     elif [ "${topology}" = "chatglm_6b" ]; then
         alpha=0.75
         model_name_or_path="THUDM/chatglm-6b"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.33
         script="run_generation_sq.py"
     elif [ "${topology}" = "falcon_7b" ]; then
         alpha=0.7
         model_name_or_path="tiiuae/falcon-7b-instruct"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
-	    pip install transformers==4.33.3
         script="run_generation_sq.py"
-    elif [ "${topology}" = "baichuan_7b" ]; then
-        alpha=0.85
-        model_name_or_path="baichuan-inc/Baichuan-7B"
-        extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
-        extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
-        extra_cmd=$extra_cmd" --trust_remote_code"
-        script="run_generation_sq.py"
-        pip install transformers==4.33
     elif [ "${topology}" = "baichuan_13b" ]; then
         alpha=0.85
-        model_name_or_path="baichuan-inc/Baichuan-13B-Base"
+        model_name_or_path="baichuan-inc/Baichuan-13B-Chat"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        extra_cmd=$extra_cmd" --_commit_hash 14d5b0e204542744900f6fb52422c6d633bdcb00"
-        pip install transformers==4.33
         script="run_generation_sq.py"
     elif [ "${topology}" = "baichuan2_7b" ]; then
         alpha=0.85
-        model_name_or_path="baichuan-inc/Baichuan2-7B-Base"
+        model_name_or_path="baichuan-inc/Baichuan2-7B-Chat"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.33
         script="run_generation_sq.py"
     elif [ "${topology}" = "baichuan2_13b" ]; then
         alpha=0.55
-        model_name_or_path="baichuan-inc/Baichuan2-13B-Base"
+        model_name_or_path="baichuan-inc/Baichuan2-13B-Chat"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        pip install transformers==4.35.2
         script="run_generation_sq.py"
     elif [ "${topology}" = "qwen_7b" ]; then
         alpha=0.9
-        model_name_or_path="Qwen/Qwen-7B"
+        model_name_or_path="Qwen/Qwen-7B-Chat"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
-        extra_cmd=$extra_cmd" --_commit_hash f7bc352f27bb1c02ee371a4576942a7d96c8bb97"
-	      pip install transformers==4.35.2
-          script="run_generation_sq.py"
+        script="run_generation_sq.py"
     elif [ "${topology}" = "mistral_7b" ]; then
         alpha=0.8
         model_name_or_path="Intel/neural-chat-7b-v3"
@@ -240,15 +220,13 @@ function run_tuning {
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
-	    pip install transformers==4.36.1
         script="run_generation_sq.py"
     elif [ "${topology}" = "phi_1_5b" ]; then
         alpha=0.5
         model_name_or_path="susnato/phi-1_5_dev"
         extra_cmd=$extra_cmd" --sq --alpha ${alpha}"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
-	    pip install transformers==4.36.1
         script="run_generation_sq.py"
     elif [ "${topology}" = "llama2_7b_gptq" ]; then
         model_name_or_path="meta-llama/Llama-2-7b-hf"

diff --git a/intel_extension_for_transformers/transformers/llm/evaluation/models.py b/intel_extension_for_transformers/transformers/llm/evaluation/models.py
@@ -166,14 +166,9 @@ def forward(
         input_bs, input_len = input_ids.shape
         if self.use_cache and past_key_values is None:
             if model_type in IPEX_OPT_LLM_SUPPORTED:
-                if model_type == "llama" and transformers.__version__ >= "4.36":
-                    past_key_values = generate_dummy_past_key_values_for_inference(
-                        config=self.config, input_bs=input_bs
-                    )
-                else:
-                    past_key_values = generate_dummy_past_key_values_for_opt_llm(
-                        config=self.config, input_bs=input_bs, num_beams=1
-                    )
+                past_key_values = generate_dummy_past_key_values_for_opt_llm(
+                    config=self.config, input_bs=input_bs, num_beams=1
+                )
             else:
                 past_key_values = generate_dummy_past_key_values_for_inference(
                     config=self.config, input_bs=input_bs

diff --git a/intel_extension_for_transformers/transformers/modeling/modeling_auto.py b/intel_extension_for_transformers/transformers/modeling/modeling_auto.py
@@ -841,8 +841,7 @@ def forward(self, input: torch.Tensor) -> tuple[torch.Tensor, None]:
                 model = model.float()
             model.eval()
             model_type = model.config.model_type.replace("_", "-")
-            if "llama" in model_type and transformers.__version__ >= "4.36.0":
-                quantization_config.ipex_opt_llm = False
+
             logger.info("Applying SmoothQuant.")
             # ipex.optimize_transformers
             if quantization_config.ipex_opt_llm is None:
@@ -851,7 +850,7 @@ def forward(self, input: torch.Tensor) -> tuple[torch.Tensor, None]:
                     logger.info(
                         "quantization_config.ipex_opt_llm set to True and ipex.optimize_transformers is used."
                     )
-                    logger.warning("The suggested transformers version is 4.35.2.")
+                    logger.warning("The suggested transformers version is 4.38.1.")
                 else:
                     quantization_config.ipex_opt_llm = False
             if quantization_config.ipex_opt_llm:
@@ -946,7 +945,7 @@ def collate_batch(batch):
                             )
 
                         last_ind.append(input_ids.shape[0] - 1)
-                        if model_type in ["bloom", "qwen"]:
+                        if model_type in ["bloom"]:
                             attention_mask = torch.ones(len(input_ids) + 1)
                             attention_mask[0] = 0
                         else:

diff --git a/intel_extension_for_transformers/transformers/utils/utility.py b/intel_extension_for_transformers/transformers/utils/utility.py
@@ -21,6 +21,7 @@
 from typing import Optional, Tuple
 from neural_compressor.utils import logger
 from neural_compressor.utils.utility import LazyImport, CpuInfo
+from intel_extension_for_transformers.tools.utils import is_ipex_available
 
 
 CONFIG_NAME = "best_configure.yaml"
@@ -36,6 +37,8 @@
 SAFE_WEIGHTS_NAME = "model.safetensors"
 SAFE_WEIGHTS_INDEX_NAME = "model.safetensors.index.json"
 
+if is_ipex_available():
+    import intel_extension_for_pytorch as ipex
 torch = LazyImport("torch")
 
 def str2bool(v):
@@ -300,8 +303,24 @@ def generate_dummy_past_key_values_for_opt_llm(config, input_bs, num_beams=1):
     ]
     return tuple(past_key_values)
 
-
-IPEX_OPT_LLM_SUPPORTED = {"gptj", "opt", "llama", "falcon", "chatglm", "baichuan"}
+IPEX_OPT_LLM_SUPPORTED_DICT = {
+    "2.2": ["gptj", "opt", "llama", "falcon", "chatglm", "baichuan", "gpt-neox"],
+    "2.3": [
+        "gptj",
+        "opt",
+        "llama",
+        "falcon",
+        "chatglm",
+        "baichuan",
+        "qwen",
+        "bloom",
+        "codegen",
+        "gptbigcode",
+        "t5",
+        "mixtral",
+        "mpt",
+    ],
+}
 
 MODEL_TYPES_REQUIRING_POSITION_IDS = {
     "codegen",
@@ -314,9 +333,32 @@ def generate_dummy_past_key_values_for_opt_llm(config, input_bs, num_beams=1):
     "llama",
     "mistral",
     "chatglm",
-    "baichuan"
 }
 
+if is_ipex_available() and ipex.__version__ == "2.2.0+cpu":
+    logger.info(
+        "ipex.llm.optimize by 2.2.0 version supported model family: {}".format(
+            ",".join(IPEX_OPT_LLM_SUPPORTED_DICT["2.2"])
+            )
+    )
+    logger.info(
+        "The recommended transformers version is 4.35.2 if you used IPEX 2.2.0 version."
+    )
+    IPEX_OPT_LLM_SUPPORTED = IPEX_OPT_LLM_SUPPORTED_DICT["2.2"]
+elif is_ipex_available() and ipex.__version__ == "2.3.0+cpu":
+    logger.info(
+        "ipex.llm.optimize by 2.3.0 version supported model family: {}".format(
+            ", ".join(IPEX_OPT_LLM_SUPPORTED_DICT["2.3"])
+        )
+    )
+    logger.info(
+        "The recommended transformers version is 4.38.1 if you used IPEX 2.3.0 version."
+    )
+    IPEX_OPT_LLM_SUPPORTED = IPEX_OPT_LLM_SUPPORTED_DICT["2.3"]
+else:
+    logger.warning("Please check the intel_extension_for_pytorch version is 2.3.0+cpu.")
+    IPEX_OPT_LLM_SUPPORTED = IPEX_OPT_LLM_SUPPORTED_DICT["2.3"]
+
 def get_example_inputs(model_config, batch_size=1, tokenizer=None, num_beams=4):
     """Generate the dummy example inputs."""
     prompt = "Welcome to use Intel Extension for Transformers."
@@ -420,7 +462,8 @@ def recover_model_from_json(fp32_model_name_or_path, json_file_path, trust_remot
         (object): quantized model
     """
     from transformers import AutoModelForCausalLM
-    user_model = AutoModelForCausalLM.from_pretrained(fp32_model_name_or_path, trust_remote_code=trust_remote_code)
+    user_model = AutoModelForCausalLM.from_pretrained(fp32_model_name_or_path,
+                                                      trust_remote_code=trust_remote_code).float()
     if user_model.config.model_type in IPEX_OPT_LLM_SUPPORTED:
         import intel_extension_for_pytorch as ipex
         qconfig = ipex.quantization.default_static_qconfig_mapping