add mx quant

docs/source/mx_quantization.md

@@ -0,0 +1,130 @@
+Microscaling Quantization
+===============
+
+1. [Introduction](#introduction)
+2. [Supported Framework Model Matrix](#supported-framework-model-matrix)
+3. [Get Started with Microscaling Quantization API](#get-start-with-microscaling-quantization-api)
+4. [Examples](#examples)
+5. [Reference](#reference)
+
+## Introduction
+
+Numerous breakthroughs have emerged across various fields, such as text analysis, language translation and chatbot technologies, fueled by the development of large language models (LLMs). Nevertheless, their increasing power comes with the challenge of explosive growth in parameters, posing obstacles for practical use. To balance memory limits and accuracy preservation for AI models, the Microscaling (MX) specification was promoted from the well-known Microsoft Floating Point (MSFP) data type [1, 2]:
+
+<table>
+  <tr>
+    <th>Format Name</th>
+    <th>Element Data type</th>
+    <th>Element Bits</th>
+    <th>Scaling Block Size</th>
+    <th>Scale Data Type</th> 
+    <th>Scale Bits</th>
+  </tr>
+  <tr>
+    <td rowspan="2">MXFP8</td>
+    <td>FP8 (E5M2)</td>
+    <td rowspan="2">8</td>
+    <td rowspan="2">32</td>
+    <td rowspan="2">E8M0</td>
+    <td rowspan="2">8</td>
+  </tr>
+  <tr>
+    <td>FP8 (E4M3)</td>
+  </tr>
+  <tr>
+    <td rowspan="2">MXFP6</td>
+    <td>FP6 (E3M2)</td>
+    <td rowspan="2">6</td>
+    <td rowspan="2">32</td>
+    <td rowspan="2">E8M0</td>
+    <td rowspan="2">8</td>
+  </tr>
+  <tr>
+    <td>FP6 (E2M3)</td>
+  </tr>
+  <tr>
+    <td>MXFP4</td>
+    <td>FP4 (E2M1)</td>
+    <td>4</td>
+    <td>32</td>
+    <td>E8M0</td> 
+    <td>8</td>
+  </tr>
+  <tr>
+    <td>MXINT8</td>
+    <td>INT8</td>
+    <td>8</td>
+    <td>32</td>
+    <td>E8M0</td> 
+    <td>8</td>
+  </tr>
+</table>
+
+
+At an equivalent accuracy level, the MX data type demonstrates the ability to occupy a smaller area and incur lower energy costs for multiply-accumulate compared to other conventional data types on the same silicon [1].
+
+Neural Compressor seamlessly applies the MX data type to post-training quantization, offering meticulously crafted recipes to empower users to quantize LLMs without sacrificing accuracy. The workflow is shown as below.
+
+<a target="_blank" href="./imgs/mx_workflow.png" text-align:left>
+    <left> 
+        <img src="./imgs/mx_workflow.png" alt="Workflow of MX Quant (source [3])" height=120> 
+    </left>
+</a>
+
+The memory and computational limits of LLMs are more severe than other general neural networks, so our exploration focuses on LLMs first. The following table shows the basic MX quantization recipes in Neural Compressor and enumerates distinctions among various data types. The MX data type replaces general float scale with powers of two to be more hardware-friendly. It adapts a granularity falling between per-channel and per-tensor to balance accuracy and memory consumption.
+
+|            | MX Format |  INT8  |  FP8  |
+|------------|--------------|------------|------------|
+|  Scale  |   $2^{exp}$   |  $\frac{MAX}{amax}$  |  $\frac{MAX}{amax}$  |
+|  Zero point  |   0 (None)   | $2^{bits - 1}$ or $-min * scale$ |   0 (None)   |
+|  Granularity  |  per-block (default blocksize is 32)   |  per-channel or per-tensor  | per-tensor  |
+
+The exponent (exp) is equal to torch.floor(torch.log2(amax)), MAX is the representation range of the data type, amax is the max absolute value of per-block tensor, and rmin is the minimum value of the per-block tensor.
+
+
+## Supported Framework Model Matrix
+
+
+<table>
+  <tr>
+    <th>Framework</th>
+    <th>Status</th>
+  </tr>
+  <tr>
+    <td>PyTorch</td>
+    <td>&#10004;</td>
+  </tr>
+  <tr>
+    <td>ONNX Runtime</td>
+    <td>&#10005;</td>
+  </tr>
+  <tr>
+    <td>TensorFlow</td>
+    <td>&#10005;</td>
+  </tr>
+</table>
+
+
+## Get Started with Microscaling Quantization API
+
+To get a model quantized with Microscaling Data Types, users can use the Microscaling Quantization API as follows.
+
+```python
+from neural_compressor.torch.quantization import MXQuantConfig, quantize
+
+quant_config = MXQuantConfig(w_dtype=args.w_dtype, act_dtype=args.act_dtype, weight_only=args.woq)
+user_model = quantize(model=user_model, quant_config=quant_config)
+```
+
+## Examples
+
+- PyTorch [huggingface models](/examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/mx)
+
+
+## Reference
+
+[1]: Darvish Rouhani, Bita, et al. "Pushing the limits of narrow precision inferencing at cloud scale with microsoft floating point." Advances in neural information processing systems 33 (2020): 10271-10281 
+
+[2]: OCP Microscaling Formats (MX) Specification
+
+[3]: Rouhani, Bita Darvish, et al. "Microscaling Data Formats for Deep Learning." arXiv preprint arXiv:2310.10537 (2023). 

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/mx/README.md

@@ -0,0 +1,6 @@
+# Run
+
+## Run WOQ MX FP4 model
+``` python
+python run_clm_no_trainer.py --model [model_name_or_id] --quantize --accuracy --tasks lambada_openai --w_dtype fp4 --woq
+```

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/mx/requirements.txt

@@ -0,0 +1,7 @@
+transformers
+torch
+sentencepiece
+neural-compressor
+intel-extension-for-transformers >= 1.4.1
+lm-eval==0.4.2
+peft

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/mx/run_clm_no_trainer.py

@@ -0,0 +1,110 @@
+import argparse
+import time
+import json
+
+parser = argparse.ArgumentParser()
+parser.add_argument(
+    "--model", nargs="?", default="EleutherAI/gpt-j-6b"
+)
+parser.add_argument(
+    "--trust_remote_code", default=True,
+    help="Transformers parameter: use the external repo")
+parser.add_argument(
+    "--revision", default=None,
+    help="Transformers parameter: set the model hub commit number")
+parser.add_argument("--quantize", action="store_true")
+# dynamic only now
+parser.add_argument("--w_dtype", type=str, default="int8", 
+                    choices=["int8", "int4", "int2", "fp8_e5m2", "fp8_e4m3", "fp6_e3m2", 
+                                                "fp6_e2m3", "fp4", "float16", "bfloat12"],
+                    help="weight data type")
+parser.add_argument("--act_dtype", type=str, default="int8", 
+                    choices=["int8", "int4", "int2", "fp8_e5m2", "fp8_e4m3", "fp6_e3m2", 
+                                                "fp6_e2m3", "fp4", "float16", "bfloat12"],
+                    help="input activation data type")
+parser.add_argument("--woq", action="store_true")
+parser.add_argument("--accuracy", action="store_true")
+parser.add_argument("--performance", action="store_true")
+parser.add_argument("--iters", default=100, type=int,
+                    help="For accuracy measurement only.")
+parser.add_argument("--batch_size", default=1, type=int,
+                    help="For accuracy measurement only.")
+parser.add_argument("--save_accuracy_path", default=None,
+                    help="Save accuracy results path.")
+parser.add_argument("--tasks", type=str, default="lambada_openai",
+                    help="tasks list for accuracy validation")
+parser.add_argument("--peft_model_id", type=str, default=None, help="model_name_or_path of peft model")
+
+args = parser.parse_args()
+
+def get_user_model():
+    from transformers import AutoModelForCausalLM, AutoModel, AutoTokenizer
+    user_model = AutoModelForCausalLM.from_pretrained(
+        args.model,
+        trust_remote_code=args.trust_remote_code,
+        revision=args.revision,
+    )
+    tokenizer = AutoTokenizer.from_pretrained(args.model, trust_remote_code=args.trust_remote_code)
+
+    if args.peft_model_id is not None:
+        from peft import PeftModel
+        user_model = PeftModel.from_pretrained(user_model, args.peft_model_id)
+
+    user_model.eval()
+    return user_model, tokenizer
+
+user_model, tokenizer = get_user_model()
+if args.quantize:
+    from neural_compressor.torch.quantization import MXQuantConfig, quantize
+    quant_config = MXQuantConfig(w_dtype=args.w_dtype, act_dtype=args.act_dtype, weight_only=args.woq)
+    user_model = quantize(model=user_model, quant_config=quant_config)
+
+
+if args.accuracy:
+    user_model.eval()
+    from intel_extension_for_transformers.transformers.llm.evaluation.lm_eval import evaluate, LMEvalParser
+    args = LMEvalParser(
+        model="hf",
+        user_model=user_model,
+        tokenizer=tokenizer,
+        batch_size=args.batch_size,
+        tasks=args.tasks,
+        device="cpu",
+    )
+    results = evaluate(args)
+    dumped = json.dumps(results, indent=2)
+    if args.save_accuracy_path:
+        with open(args.save_accuracy_path, "w") as f:
+            f.write(dumped)
+    for task_name in args.tasks:
+        if task_name == "wikitext":
+            acc = results["results"][task_name]["word_perplexity"]
+        else:
+            acc = results["results"][task_name]["acc"]
+    print("Accuracy: %.5f" % acc)
+    print('Batch size = %d' % args.batch_size)
+
+if args.performance:
+    user_model.eval()
+    from intel_extension_for_transformers.llm.evaluation.lm_eval import evaluate
+    import time
+    samples = args.iters * args.batch_size
+    start = time.time()
+    results = evaluate(
+        model="hf",
+        tokenizer=tokenizer,
+        user_model=user_model,
+        batch_size=args.batch_size,
+        tasks=args.tasks,
+        limit=samples,
+    )
+    end = time.time()
+    for task_name in args.tasks:
+        if task_name == "wikitext":
+            acc = results["results"][task_name]["word_perplexity"]
+        else:
+            acc = results["results"][task_name]["acc"]
+    print("Accuracy: %.5f" % acc)
+    print('Throughput: %.3f samples/sec' % (samples / (end - start)))
+    print('Latency: %.3f ms' % ((end - start)*1000 / samples))
+    print('Batch size = %d' % args.batch_size)

neural_compressor/common/utils/constants.py

@@ -36,6 +36,7 @@
 TEQ = "teq"  # pragma: no cover
 AUTOROUND = "autoround"
 FP8_QUANT = "fp8_quant"
+MX_QUANT = "mx_quant"
 MIX_PRECISION = "mix_precision"
 
 # options

neural_compressor/torch/algorithms/mx_quant/__init__.py

@@ -0,0 +1,15 @@
+# Copyright (c) 2024 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# pylint:disable=import-error