[Research] Custom Deepseek Routing

examples/quantization_w4a16/gptj_example.py

@@ -0,0 +1,55 @@
+from datasets import load_dataset
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+from llmcompressor.modifiers.quantization import GPTQModifier
+from llmcompressor.modifiers.smoothquant.base import SmoothQuantModifier
+from llmcompressor.transformers import oneshot
+
+# Select model and load it.
+MODEL_ID = "EleutherAI/gpt-j-6B"
+
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    device_map="auto",
+    torch_dtype="auto",
+)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+
+# Select calibration dataset.
+DATASET_ID = "HuggingFaceH4/ultrachat_200k"
+DATASET_SPLIT = "train_sft"
+
+# Select number of samples. 512 samples is a good place to start.
+# Increasing the number of samples can improve accuracy.
+NUM_CALIBRATION_SAMPLES = 512
+MAX_SEQUENCE_LENGTH = 2048
+
+# Configure the quantization algorithm to run.
+#   * quantize the weights to 4 bit with GPTQ with a group size 128
+recipe = [
+    SmoothQuantModifier(smoothing_strength=0.8),
+    GPTQModifier(targets="Linear", scheme="W4A16", ignore=["lm_head"]),
+]
+
+# Apply algorithms.
+oneshot(
+    model=model,
+    dataset="ultrachat-200k",
+    splits={"calibration": f"train_sft[:{NUM_CALIBRATION_SAMPLES}]"},
+    recipe=recipe,
+    max_seq_length=MAX_SEQUENCE_LENGTH,
+    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
+)
+
+# Confirm generations of the quantized model look sane.
+print("\n\n")
+print("========== SAMPLE GENERATION ==============")
+input_ids = tokenizer("Hello my name is", return_tensors="pt").input_ids.to("cuda")
+output = model.generate(input_ids, max_new_tokens=100)
+print(tokenizer.decode(output[0]))
+print("==========================================\n\n")
+
+# Save to disk compressed.
+SAVE_DIR = MODEL_ID.split("/")[1] + "-W4A16-G128"
+model.save_pretrained(SAVE_DIR, save_compressed=True)
+tokenizer.save_pretrained(SAVE_DIR)

examples/quantizing_moe/deepseek_moe_w4a16.py

@@ -1,29 +1,40 @@
 import torch
 from datasets import load_dataset
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoTokenizer
 
 from llmcompressor.transformers import oneshot
 from llmcompressor.transformers.compression.helpers import calculate_offload_device_map
+from llmcompressor.transformers.tracing import TraceableDeepseekV2ForCausalLM
+from llmcompressor.transformers.tracing.deepseek_v2.configuration_deepseek import (
+    DeepseekV2Config,
+)
 
 # NOTE: transformers 4.48.0 has an import error with DeepSeek.
 # Please consider either downgrading your transformers version to a
 # previous version or upgrading to a version where this bug is fixed
 
 # select a Mixture of Experts model for quantization
-MODEL_ID = "deepseek-ai/DeepSeek-V2.5"
+MODEL_ID = "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct"
 
 # adjust based off number of desired GPUs
 # if not enough memory is available, some layers will automatically be offlaoded to cpu
 device_map = calculate_offload_device_map(
     MODEL_ID,
     reserve_for_hessians=True,
-    num_gpus=2,
+    num_gpus=1,
     torch_dtype=torch.bfloat16,
     trust_remote_code=True,
 )
 
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL_ID, device_map=device_map, torch_dtype=torch.bfloat16, trust_remote_code=True
+# model = AutoModelForCausalLM.from_pretrained(
+config = DeepseekV2Config.from_pretrained(MODEL_ID)
+config.moe_top_k_activation = True
+model = TraceableDeepseekV2ForCausalLM.from_pretrained(
+    MODEL_ID,
+    device_map=device_map,
+    torch_dtype=torch.bfloat16,
+    trust_remote_code=True,
+    config=config,
 )
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 
@@ -91,35 +102,35 @@ def tokenize(sample):
 print("==========================================")
 
 
-# Run the model on vLLM
-try:
-    from vllm import LLM, SamplingParams
-
-    vllm_installed = True
-except ImportError:
-    vllm_installed = False
-
-if vllm_installed:
-    print("vLLM installed, running using vLLM")
-    sampling_params = SamplingParams(temperature=0.80, top_p=0.95)
-    llm = LLM(
-        model=SAVE_DIR,
-        tensor_parallel_size=2,
-        trust_remote_code=True,
-        max_model_len=1042,
-        dtype=torch.half,
-    )
-    prompts = [
-        "The capital of France is",
-        "The president of the US is",
-        "My name is",
-    ]
-
-    outputs = llm.generate(prompts, sampling_params)
-    print("================= vLLM GENERATION ======================")
-    for output in outputs:
-        assert output
-        prompt = output.prompt
-        generated_text = output.outputs[0].text
-        print("PROMPT", prompt)
-        print("GENERATED TEXT", generated_text)
+# # Run the model on vLLM
+# try:
+#     from vllm import LLM, SamplingParams
+
+#     vllm_installed = True
+# except ImportError:
+#     vllm_installed = False
+
+# if vllm_installed:
+#     print("vLLM installed, running using vLLM")
+#     sampling_params = SamplingParams(temperature=0.80, top_p=0.95)
+#     llm = LLM(
+#         model=SAVE_DIR,
+#         tensor_parallel_size=2,
+#         trust_remote_code=True,
+#         max_model_len=1042,
+#         dtype=torch.half,
+#     )
+#     prompts = [
+#         "The capital of France is",
+#         "The president of the US is",
+#         "My name is",
+#     ]
+
+#     outputs = llm.generate(prompts, sampling_params)
+#     print("================= vLLM GENERATION ======================")
+#     for output in outputs:
+#         assert output
+#         prompt = output.prompt
+#         generated_text = output.outputs[0].text
+#         print("PROMPT", prompt)
+#         print("GENERATED TEXT", generated_text)

src/llmcompressor/transformers/tracing/__init__.py

@@ -5,9 +5,13 @@
 from .mllama import (
     MllamaForConditionalGeneration as TraceableMllamaForConditionalGeneration,
 )
+from .deepseek_v2.modeling_deepseek import (
+    DeepseekV2ForCausalLM as TraceableDeepseekV2ForCausalLM
+)
 
 __all__ = [
     "TraceableLlavaForConditionalGeneration",
     "TraceableMllamaForConditionalGeneration",
     "TraceableMistralForCausalLM",
+    "TraceableDeepseekV2ForCausalLM",
 ]

src/llmcompressor/transformers/tracing/deepseek_v2/configuration_deepseek.py

@@ -0,0 +1,210 @@
+# flake8: noqa
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+DEEPSEEK_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+class DeepseekV2Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`DeepseekV2Model`]. It is used to instantiate an DeepSeek
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the DeepSeek-V2.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 102400):
+            Vocabulary size of the Deep model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`DeepseekV2Model`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 11008):
+            Dimension of the MLP representations.
+        moe_intermediate_size (`int`, *optional*, defaults to 1407):
+            Dimension of the MoE representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        n_shared_experts (`int`, *optional*, defaults to None):
+            Number of shared experts, None means dense model.
+        n_routed_experts (`int`, *optional*, defaults to None):
+            Number of routed experts, None means dense model.
+        routed_scaling_factor (`float`, *optional*, defaults to 1.0):
+            Scaling factor or routed experts.
+        topk_method (`str`, *optional*, defaults to `gready`):
+            Topk method used in routed gate.
+        n_group (`int`, *optional*, defaults to None):
+            Number of groups for routed experts.
+        topk_group (`int`, *optional*, defaults to None):
+            Number of selected groups for each token(for each token, ensuring the selected experts is only within `topk_group` groups).
+        num_experts_per_tok (`int`, *optional*, defaults to None):
+            Number of selected experts, None means dense model.
+        moe_layer_freq (`int`, *optional*, defaults to 1):
+            The frequency of the MoE layer: one expert layer for every `moe_layer_freq - 1` dense layers.
+        first_k_dense_replace (`int`, *optional*, defaults to 0):
+            Number of dense layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
+                                                            \--k dense layers--/
+        norm_topk_prob (`bool`, *optional*, defaults to False):
+            Whether to normalize the weights of the routed experts.
+        scoring_func (`str`, *optional*, defaults to 'softmax'):
+            Method of computing expert weights.
+        aux_loss_alpha (`float`, *optional*, defaults to 0.001):
+            Auxiliary loss weight coefficient.
+        seq_aux = (`bool`, *optional*, defaults to True):
+            Whether to compute the auxiliary loss for each individual sample.
+        num_key_value_heads (`int`, *optional*):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
+            `num_attention_heads`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 2048):
+            The maximum sequence length that this model might ever be used with.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            End of stream token id.
+        pretraining_tp (`int`, *optional*, defaults to 1):
+            Experimental feature. Tensor parallelism rank used during pretraining. Please refer to [this
+            document](https://huggingface.co/docs/transformers/parallelism) to understand more about it. This value is
+            necessary to ensure exact reproducibility of the pretraining results. Please refer to [this
+            issue](https://github.com/pytorch/pytorch/issues/76232).
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        rope_scaling (`Dict`, *optional*):
+            Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
+            strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
+            `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
+            `max_position_embeddings` to the expected new maximum.
+        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output projection layers during self-attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+
+    ```python
+    >>> from transformers import DeepseekV2Model, DeepseekV2Config
+
+    >>> # Initializing a Deepseek-V2 style configuration
+    >>> configuration = DeepseekV2Config()
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "deepseek_v2"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=102400,
+        hidden_size=4096,
+        intermediate_size=11008,
+        moe_intermediate_size = 1407,
+        num_hidden_layers=30,
+        num_attention_heads=32,
+        num_key_value_heads=32,
+        n_shared_experts = None,
+        n_routed_experts = None,
+        ep_size = 1,
+        routed_scaling_factor = 1.0,
+        kv_lora_rank = 512,
+        q_lora_rank = 1536,
+        qk_rope_head_dim = 64,
+        v_head_dim = 128,
+        qk_nope_head_dim = 128,
+        topk_method = 'gready',
+        n_group = None,
+        topk_group = None,
+        num_experts_per_tok = None,
+        moe_layer_freq = 1,
+        first_k_dense_replace = 0,
+        norm_topk_prob = False,
+        scoring_func = 'softmax',
+        aux_loss_alpha = 0.001,
+        seq_aux = True,
+        hidden_act="silu",
+        max_position_embeddings=2048,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=100000,
+        eos_token_id=100001,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        attention_dropout=0.0,
+        # TRACING: add calibration options
+        moe_top_k_activation=True,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.moe_intermediate_size = moe_intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.n_shared_experts = n_shared_experts
+        self.n_routed_experts = n_routed_experts
+        self.ep_size = ep_size
+        self.routed_scaling_factor = routed_scaling_factor
+        self.kv_lora_rank = kv_lora_rank
+        self.q_lora_rank = q_lora_rank
+        self.qk_rope_head_dim = qk_rope_head_dim
+        self.v_head_dim = v_head_dim
+        self.qk_nope_head_dim = qk_nope_head_dim
+        self.topk_method = topk_method
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.num_experts_per_tok = num_experts_per_tok
+        self.moe_layer_freq = moe_layer_freq
+        self.first_k_dense_replace = first_k_dense_replace
+        self.norm_topk_prob = norm_topk_prob
+        self.scoring_func = scoring_func
+        self.aux_loss_alpha = aux_loss_alpha
+        self.seq_aux = seq_aux
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.pretraining_tp = pretraining_tp
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        self.moe_top_k_activation = moe_top_k_activation
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )