vllm.model_executor.model_loader.neuron

Utilities for selecting and loading Neuron models in transformers-neuronx framework.

TORCH_DTYPE_TO_NEURON_AMP module-attribute

TORCH_DTYPE_TO_NEURON_AMP = {
    "auto": "f32",
    "half": "f16",
    "float16": "f16",
    "bfloat16": "bf16",
    "float": "f32",
    "float32": "f32",
    float16: "f16",
    bfloat16: "bf16",
    float32: "f32",
}

_NEURON_SUPPORTED_MODELS module-attribute

_NEURON_SUPPORTED_MODELS: dict[
    str, tuple[str, str, str]
] = {
    "LlamaForCausalLM": (
        "transformers_neuronx.llama.model",
        "LlamaForSampling",
        "LlamaForCausalLM",
    ),
    "MistralForCausalLM": (
        "transformers_neuronx.mistral.model",
        "MistralForSampling",
        "MistralForCausalLM",
    ),
}

NeuronCausalLM

Bases: Module

Source code in vllm/model_executor/model_loader/neuron.py

class NeuronCausalLM(nn.Module):

    def __init__(self,
                 config: PretrainedConfig,
                 on_device_sampling_disabled: bool = False) -> None:
        super().__init__()
        self.config = config
        self.logits_processor = LogitsProcessor(config.vocab_size,
                                                logits_as_input=True)

        self.on_device_sampling_disabled = on_device_sampling_disabled
        if self.on_device_sampling_disabled:
            # Use default sampler
            self.sampler = Sampler()

        # Lazy initialized
        self.model: nn.Module

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        input_block_ids: torch.Tensor,
    ) -> torch.Tensor:
        logits = self.model(input_ids,
                            cache_ids=positions,
                            start_ids=input_block_ids)
        return logits

    def compute_logits(self, hidden_states: torch.Tensor,
                       sampling_metadata: SamplingMetadata) -> torch.Tensor:
        logits = self.logits_processor(None, hidden_states, sampling_metadata)
        return logits

    def sample(
        self,
        logits: torch.Tensor,
        sampling_metadata: SamplingMetadata,
    ) -> Optional[SamplerOutput]:

        if self.on_device_sampling_disabled:
            next_tokens = self.sampler(logits, sampling_metadata)
            return next_tokens

        # On-device sampling outputs the token ids directly.
        sampled_token_ids = logits.flatten()
        next_tokens = []
        sample_idx = 0
        for seq_group in sampling_metadata.seq_groups:
            samples = []
            for seq_id in seq_group.seq_ids:
                token_id = sampled_token_ids[sample_idx].item()
                samples.append(
                    SequenceOutput(parent_seq_id=seq_id,
                                   output_token=token_id,
                                   logprobs={token_id: Logprob(token_id)}))
                sample_idx += 1
            next_tokens.append(
                CompletionSequenceGroupOutput(samples=samples,
                                              prompt_logprobs=None))

        return SamplerOutput(outputs=next_tokens)

    def load_weights(self, model_name_or_path: str, **kwargs):
        arch = _get_model_architecture(self.config)
        neuronx_module_path, neuronx_model_cls_name, hf_model_cls_name = (
            _NEURON_SUPPORTED_MODELS[arch])
        neuronx_module = importlib.import_module(neuronx_module_path)
        neuronx_model_cls = getattr(neuronx_module, neuronx_model_cls_name)

        self.model = neuronx_model_cls.from_pretrained(model_name_or_path,
                                                       **kwargs)
        self.model.to_neuron()

config instance-attribute

config = config

logits_processor instance-attribute

logits_processor = LogitsProcessor(
    vocab_size, logits_as_input=True
)

model instance-attribute

model: Module

on_device_sampling_disabled instance-attribute

on_device_sampling_disabled = on_device_sampling_disabled

sampler instance-attribute

sampler = Sampler()

__init__

__init__(
    config: PretrainedConfig,
    on_device_sampling_disabled: bool = False,
) -> None

Source code in vllm/model_executor/model_loader/neuron.py

def __init__(self,
             config: PretrainedConfig,
             on_device_sampling_disabled: bool = False) -> None:
    super().__init__()
    self.config = config
    self.logits_processor = LogitsProcessor(config.vocab_size,
                                            logits_as_input=True)

    self.on_device_sampling_disabled = on_device_sampling_disabled
    if self.on_device_sampling_disabled:
        # Use default sampler
        self.sampler = Sampler()

    # Lazy initialized
    self.model: nn.Module

compute_logits

compute_logits(
    hidden_states: Tensor,
    sampling_metadata: SamplingMetadata,
) -> Tensor

Source code in vllm/model_executor/model_loader/neuron.py

def compute_logits(self, hidden_states: torch.Tensor,
                   sampling_metadata: SamplingMetadata) -> torch.Tensor:
    logits = self.logits_processor(None, hidden_states, sampling_metadata)
    return logits

forward

forward(
    input_ids: Tensor,
    positions: Tensor,
    input_block_ids: Tensor,
) -> Tensor

Source code in vllm/model_executor/model_loader/neuron.py

def forward(
    self,
    input_ids: torch.Tensor,
    positions: torch.Tensor,
    input_block_ids: torch.Tensor,
) -> torch.Tensor:
    logits = self.model(input_ids,
                        cache_ids=positions,
                        start_ids=input_block_ids)
    return logits

load_weights

load_weights(model_name_or_path: str, **kwargs)

Source code in vllm/model_executor/model_loader/neuron.py

def load_weights(self, model_name_or_path: str, **kwargs):
    arch = _get_model_architecture(self.config)
    neuronx_module_path, neuronx_model_cls_name, hf_model_cls_name = (
        _NEURON_SUPPORTED_MODELS[arch])
    neuronx_module = importlib.import_module(neuronx_module_path)
    neuronx_model_cls = getattr(neuronx_module, neuronx_model_cls_name)

    self.model = neuronx_model_cls.from_pretrained(model_name_or_path,
                                                   **kwargs)
    self.model.to_neuron()

sample

sample(
    logits: Tensor, sampling_metadata: SamplingMetadata
) -> Optional[SamplerOutput]

Source code in vllm/model_executor/model_loader/neuron.py

def sample(
    self,
    logits: torch.Tensor,
    sampling_metadata: SamplingMetadata,
) -> Optional[SamplerOutput]:

    if self.on_device_sampling_disabled:
        next_tokens = self.sampler(logits, sampling_metadata)
        return next_tokens

    # On-device sampling outputs the token ids directly.
    sampled_token_ids = logits.flatten()
    next_tokens = []
    sample_idx = 0
    for seq_group in sampling_metadata.seq_groups:
        samples = []
        for seq_id in seq_group.seq_ids:
            token_id = sampled_token_ids[sample_idx].item()
            samples.append(
                SequenceOutput(parent_seq_id=seq_id,
                               output_token=token_id,
                               logprobs={token_id: Logprob(token_id)}))
            sample_idx += 1
        next_tokens.append(
            CompletionSequenceGroupOutput(samples=samples,
                                          prompt_logprobs=None))

    return SamplerOutput(outputs=next_tokens)

NeuronSpeculationCausalLM

Bases: Module

A Neuron-optimized causal language model with speculative decoding.

Source code in vllm/model_executor/model_loader/neuron.py

class NeuronSpeculationCausalLM(nn.Module):
    """A Neuron-optimized causal language model with speculative decoding."""

    SPECULATION_TERMINATION_ID = -1

    def __init__(self, speculation_model) -> None:
        super().__init__()
        self.model = speculation_model

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        input_block_ids: torch.Tensor,
    ) -> torch.Tensor:
        tokens, counts = self.model.speculative_iteration(
            input_ids, positions, input_block_ids)

        # Mark the end of accepted speculative tokens for each sequence with the
        # speculation termination id.
        batch_size, steps = tokens.shape
        mask = torch.arange(steps).expand(batch_size, -1) >= counts
        tokens[mask] = self.SPECULATION_TERMINATION_ID

        return tokens

    def sample(
        self,
        logits: torch.Tensor,
        sampling_metadata: SamplingMetadata,
    ) -> Optional[list[SamplerOutput]]:
        batch_size, num_steps = logits.shape
        seq_ids = [
            seq_id for sg in sampling_metadata.seq_groups
            for seq_id in sg.seq_ids
        ]
        # Organize input tensors by step instead of by sequence.
        accepted_token_ids_by_step = logits.transpose(0, 1)
        accepted_token_ids_by_step = accepted_token_ids_by_step.tolist()

        sampler_output_list = []
        for step_index in range(num_steps):
            if all(token_id == self.SPECULATION_TERMINATION_ID
                   for token_id in accepted_token_ids_by_step[step_index]):
                break
            step_output_token_ids = []
            for sequence_index in range(batch_size):
                token_id = accepted_token_ids_by_step[step_index][
                    sequence_index]
                step_output_token_ids.append(
                    CompletionSequenceGroupOutput(samples=[
                        SequenceOutput(parent_seq_id=seq_ids[sequence_index],
                                       output_token=token_id,
                                       logprobs={token_id: Logprob(token_id)})
                    ],
                                                  prompt_logprobs=None))
            sampler_output_list.append(
                SamplerOutput(outputs=step_output_token_ids))
        return sampler_output_list

SPECULATION_TERMINATION_ID class-attribute instance-attribute

SPECULATION_TERMINATION_ID = -1

model instance-attribute

model = speculation_model

__init__

__init__(speculation_model) -> None

Source code in vllm/model_executor/model_loader/neuron.py

def __init__(self, speculation_model) -> None:
    super().__init__()
    self.model = speculation_model

forward

forward(
    input_ids: Tensor,
    positions: Tensor,
    input_block_ids: Tensor,
) -> Tensor

Source code in vllm/model_executor/model_loader/neuron.py

def forward(
    self,
    input_ids: torch.Tensor,
    positions: torch.Tensor,
    input_block_ids: torch.Tensor,
) -> torch.Tensor:
    tokens, counts = self.model.speculative_iteration(
        input_ids, positions, input_block_ids)

    # Mark the end of accepted speculative tokens for each sequence with the
    # speculation termination id.
    batch_size, steps = tokens.shape
    mask = torch.arange(steps).expand(batch_size, -1) >= counts
    tokens[mask] = self.SPECULATION_TERMINATION_ID

    return tokens

sample

sample(
    logits: Tensor, sampling_metadata: SamplingMetadata
) -> Optional[list[SamplerOutput]]

Source code in vllm/model_executor/model_loader/neuron.py

def sample(
    self,
    logits: torch.Tensor,
    sampling_metadata: SamplingMetadata,
) -> Optional[list[SamplerOutput]]:
    batch_size, num_steps = logits.shape
    seq_ids = [
        seq_id for sg in sampling_metadata.seq_groups
        for seq_id in sg.seq_ids
    ]
    # Organize input tensors by step instead of by sequence.
    accepted_token_ids_by_step = logits.transpose(0, 1)
    accepted_token_ids_by_step = accepted_token_ids_by_step.tolist()

    sampler_output_list = []
    for step_index in range(num_steps):
        if all(token_id == self.SPECULATION_TERMINATION_ID
               for token_id in accepted_token_ids_by_step[step_index]):
            break
        step_output_token_ids = []
        for sequence_index in range(batch_size):
            token_id = accepted_token_ids_by_step[step_index][
                sequence_index]
            step_output_token_ids.append(
                CompletionSequenceGroupOutput(samples=[
                    SequenceOutput(parent_seq_id=seq_ids[sequence_index],
                                   output_token=token_id,
                                   logprobs={token_id: Logprob(token_id)})
                ],
                                              prompt_logprobs=None))
        sampler_output_list.append(
            SamplerOutput(outputs=step_output_token_ids))
    return sampler_output_list

_get_buckets

_get_buckets(
    env: str, default_value: list[int]
) -> list[int]

Source code in vllm/model_executor/model_loader/neuron.py

def _get_buckets(env: str, default_value: list[int]) -> list[int]:
    env_value = os.getenv(env)
    if env_value is None:
        return default_value
    buckets_remove_empty = filter(
        lambda x: x is not None and len(x.strip()) > 0, env_value.split(","))
    buckets_int = map(int, buckets_remove_empty)
    buckets_list = list(buckets_int)
    return buckets_list

_get_default_neuron_config

_get_default_neuron_config(
    model_config: ModelConfig,
    parallel_config: ParallelConfig,
    scheduler_config: SchedulerConfig,
)

Generate a neuron config based on vllm config args.

Source code in vllm/model_executor/model_loader/neuron.py

def _get_default_neuron_config(model_config: ModelConfig,
                               parallel_config: ParallelConfig,
                               scheduler_config: SchedulerConfig):
    """Generate a neuron config based on vllm config args."""
    from transformers_neuronx.config import ContinuousBatchingConfig
    from transformers_neuronx.constants import LAYOUT_BSH

    continuous_batching_config = ContinuousBatchingConfig(
        batch_size_for_shared_caches=scheduler_config.max_num_seqs)
    quant_config = dict(
        dequant_dtype=TORCH_DTYPE_TO_NEURON_AMP[model_config.dtype],
        quantize_method="vector_dynamic")
    neuron_quantization_config_builder = lambda quant: get_quantization_config(
        quant).from_config(quant_config).get_quant_method(None, "")
    # TODO: Add Paged attention config to the default neuron arguments.
    default_neuron_args = dict(
        collectives_layout=LAYOUT_BSH,
        attention_layout=LAYOUT_BSH,
        fuse_qkv=True,
        quant=neuron_quantization_config_builder(model_config.quantization)
        if model_config.quantization else None,
        continuous_batching=continuous_batching_config,
        weight_tiling=bool(model_config.quantization),
        on_device_generation=_get_neuron_on_device_generation_config(
            model_config))
    return default_neuron_args

_get_default_neuron_config_for_speculation

_get_default_neuron_config_for_speculation(
    model_config: ModelConfig,
    parallel_config: ParallelConfig,
    scheduler_config: SchedulerConfig,
)

Generate a neuron config for speculative decoding based on vllm config args.

Source code in vllm/model_executor/model_loader/neuron.py

def _get_default_neuron_config_for_speculation(
        model_config: ModelConfig, parallel_config: ParallelConfig,
        scheduler_config: SchedulerConfig):
    """Generate a neuron config for speculative decoding based on
    vllm config args."""
    from transformers_neuronx.config import ContinuousBatchingConfig
    from transformers_neuronx.constants import LAYOUT_BSH

    continuous_batching_config = ContinuousBatchingConfig(
        batch_size_for_shared_caches=scheduler_config.max_num_seqs)

    default_neuron_args = dict(collectives_layout=LAYOUT_BSH,
                               attention_layout=LAYOUT_BSH,
                               fuse_qkv=True,
                               on_device_embedding=True,
                               continuous_batching=continuous_batching_config,
                               on_device_generation=copy.deepcopy(
                                   model_config.neuron_sampling_params))
    return default_neuron_args

_get_model_architecture

_get_model_architecture(config: PretrainedConfig) -> str

Source code in vllm/model_executor/model_loader/neuron.py

def _get_model_architecture(config: PretrainedConfig) -> str:
    architectures = getattr(config, "architectures", [])
    for arch in architectures:
        if arch in _NEURON_SUPPORTED_MODELS:
            return arch
    raise ValueError(
        f"Model architectures {architectures} are not supported on Neuron "
        f"for now. Supported architectures: "
        f"{list(_NEURON_SUPPORTED_MODELS.keys())}")

_get_neuron_config_after_override

_get_neuron_config_after_override(
    default_neuron_config, overridden_neuron_config
)

Source code in vllm/model_executor/model_loader/neuron.py

def _get_neuron_config_after_override(default_neuron_config,
                                      overridden_neuron_config):
    from transformers_neuronx.config import (ContinuousBatchingConfig,
                                             GenerationConfig,
                                             KVCacheQuantizationConfig,
                                             NeuronConfig, QuantizationConfig,
                                             SparseAttnConfig)

    sparse_attn = overridden_neuron_config.pop("sparse_attn", {})
    if sparse_attn:
        overridden_neuron_config["sparse_attn"] = SparseAttnConfig(
            **sparse_attn)

    kv_cache_quant = overridden_neuron_config.pop("kv_cache_quant", {})
    if kv_cache_quant:
        overridden_neuron_config["kv_cache_quant"] = KVCacheQuantizationConfig(
            **kv_cache_quant)

    continuous_batching = overridden_neuron_config.pop("continuous_batching",
                                                       {})
    if continuous_batching:
        overridden_neuron_config[
            "continuous_batching"] = ContinuousBatchingConfig(
                **continuous_batching)

    quant = overridden_neuron_config.pop("quant", {})
    if quant:
        overridden_neuron_config["quant"] = QuantizationConfig(**quant)

    on_device_generation = overridden_neuron_config.pop(
        "on_device_generation", {})
    if on_device_generation:
        overridden_neuron_config["on_device_generation"] = GenerationConfig(
            **on_device_generation)
    default_neuron_config.update(overridden_neuron_config)
    return NeuronConfig(**default_neuron_config)

_get_neuron_on_device_generation_config

_get_neuron_on_device_generation_config(
    model_config: ModelConfig,
)

Source code in vllm/model_executor/model_loader/neuron.py

def _get_neuron_on_device_generation_config(model_config: ModelConfig):
    if not _is_neuron_on_device_sampling_disabled(model_config):
        return copy.deepcopy(model_config.neuron_sampling_params)
    return None

_is_neuron_on_device_sampling_disabled

_is_neuron_on_device_sampling_disabled(
    model_config: ModelConfig,
) -> bool

Source code in vllm/model_executor/model_loader/neuron.py

def _is_neuron_on_device_sampling_disabled(model_config: ModelConfig) -> bool:
    return not getattr(model_config, "neuron_sampling_params", None)

get_neuron_eagle_speculation_model

get_neuron_eagle_speculation_model(
    model_config: ModelConfig,
    parallel_config: ParallelConfig,
    scheduler_config: SchedulerConfig,
    speculation_config: SpeculativeConfig,
)

Initializes a neuron-optimized EAGLE speculation model for inference.

Source code in vllm/model_executor/model_loader/neuron.py

def get_neuron_eagle_speculation_model(model_config: ModelConfig,
                                       parallel_config: ParallelConfig,
                                       scheduler_config: SchedulerConfig,
                                       speculation_config: SpeculativeConfig):
    """Initializes a neuron-optimized EAGLE speculation model for inference."""
    from transformers_neuronx.eagle_speculation import EagleSpeculativeDecoder

    # Create target model instance.
    target_model = NeuronCausalLM(model_config.hf_config)

    default_neuron_config_args = _get_default_neuron_config_for_speculation(
        model_config, parallel_config, scheduler_config)
    default_neuron_config_args['is_eagle_target'] = True
    neuron_config = _get_neuron_config_after_override(
        default_neuron_config_args, model_config.override_neuron_config)

    context_length_estimates = _get_buckets("NEURON_CONTEXT_LENGTH_BUCKETS",
                                            [scheduler_config.max_model_len])
    n_positions = _get_buckets("NEURON_TOKEN_GEN_BUCKETS",
                               [scheduler_config.max_model_len])

    target_model.load_weights(
        model_config.model,
        tp_degree=parallel_config.tensor_parallel_size,
        amp=TORCH_DTYPE_TO_NEURON_AMP[model_config.dtype],
        neuron_config=neuron_config,
        context_length_estimate=context_length_estimates,
        n_positions=n_positions,
        batch_size=scheduler_config.max_num_seqs)

    target_model.eval()

    # Create draft model instance.
    draft_model = NeuronCausalLM(
        speculation_config.draft_model_config.hf_config)

    default_draft_neuron_config_args = (
        _get_default_neuron_config_for_speculation(
            speculation_config.draft_model_config, parallel_config,
            scheduler_config))
    default_draft_neuron_config_args['is_eagle_draft'] = True
    default_draft_neuron_config_args['has_pre_attention_norm'] = False
    draft_neuron_config = _get_neuron_config_after_override(
        default_draft_neuron_config_args,
        speculation_config.draft_model_config.override_neuron_config)

    draft_model.load_weights(speculation_config.draft_model_config.model,
                             tp_degree=speculation_config.
                             draft_parallel_config.tensor_parallel_size,
                             amp=TORCH_DTYPE_TO_NEURON_AMP[
                                 speculation_config.draft_model_config.dtype],
                             neuron_config=draft_neuron_config,
                             context_length_estimate=context_length_estimates,
                             n_positions=n_positions,
                             batch_size=scheduler_config.max_num_seqs)

    draft_model.eval()

    token_tree: dict[int, list[int]] = ast.literal_eval(
        speculation_config.speculative_token_tree)

    speculation_model = EagleSpeculativeDecoder(draft_model.model,
                                                target_model.model,
                                                token_tree=token_tree)
    speculation_model.to_neuron()

    return NeuronSpeculationCausalLM(speculation_model)

get_neuron_model

get_neuron_model(
    model_config: ModelConfig,
    parallel_config: ParallelConfig,
    scheduler_config: SchedulerConfig,
) -> Module

Initializes a neuron-optimized model for inference.

Source code in vllm/model_executor/model_loader/neuron.py

def get_neuron_model(model_config: ModelConfig,
                     parallel_config: ParallelConfig,
                     scheduler_config: SchedulerConfig) -> nn.Module:
    """Initializes a neuron-optimized model for inference."""
    # Create a model instance.
    model = NeuronCausalLM(
        model_config.hf_config,
        _is_neuron_on_device_sampling_disabled(model_config))

    default_neuron_config_args = _get_default_neuron_config(
        model_config, parallel_config, scheduler_config)

    neuron_config = _get_neuron_config_after_override(
        default_neuron_config_args, model_config.override_neuron_config)

    context_length_estimates = _get_buckets("NEURON_CONTEXT_LENGTH_BUCKETS",
                                            [scheduler_config.max_model_len])
    n_positions = _get_buckets("NEURON_TOKEN_GEN_BUCKETS",
                               [scheduler_config.max_model_len])

    model.load_weights(model_config.model,
                       tp_degree=parallel_config.tensor_parallel_size,
                       amp=TORCH_DTYPE_TO_NEURON_AMP[model_config.dtype],
                       neuron_config=neuron_config,
                       context_length_estimate=context_length_estimates,
                       n_positions=n_positions,
                       batch_size=scheduler_config.max_num_seqs)

    return model.eval()

get_neuron_speculation_model

get_neuron_speculation_model(
    model_config: ModelConfig,
    parallel_config: ParallelConfig,
    scheduler_config: SchedulerConfig,
    speculation_config: SpeculativeConfig,
)

Initializes a neuron-optimized speculation model for inference.

This method is only applicable for speculation with a standalone draft model

Source code in vllm/model_executor/model_loader/neuron.py

def get_neuron_speculation_model(model_config: ModelConfig,
                                 parallel_config: ParallelConfig,
                                 scheduler_config: SchedulerConfig,
                                 speculation_config: SpeculativeConfig):
    """Initializes a neuron-optimized speculation model for inference.

    This method is only applicable for speculation with a standalone draft model
    """
    from transformers_neuronx.fused_speculation import FusedSpeculativeDecoder

    # For Eagle SD, we need to pass in additional parameters in neuron config.
    is_eagle = getattr(speculation_config.draft_model_config.hf_config,
                       "is_eagle", False)

    # Create target model instance.
    target_model = NeuronCausalLM(model_config.hf_config)

    default_neuron_config_args = _get_default_neuron_config_for_speculation(
        model_config, parallel_config, scheduler_config)
    if is_eagle:
        default_neuron_config_args['is_eagle_target'] = True

    neuron_config = _get_neuron_config_after_override(
        default_neuron_config_args, model_config.override_neuron_config)

    context_length_estimates = _get_buckets("NEURON_CONTEXT_LENGTH_BUCKETS",
                                            [scheduler_config.max_model_len])
    n_positions = _get_buckets("NEURON_TOKEN_GEN_BUCKETS",
                               [scheduler_config.max_model_len])

    target_model.load_weights(
        model_config.model,
        tp_degree=parallel_config.tensor_parallel_size,
        amp=TORCH_DTYPE_TO_NEURON_AMP[model_config.dtype],
        neuron_config=neuron_config,
        context_length_estimate=context_length_estimates,
        n_positions=n_positions,
        batch_size=scheduler_config.max_num_seqs)

    target_model.eval()

    # Create draft model instance.
    draft_model = NeuronCausalLM(
        speculation_config.draft_model_config.hf_config)

    default_draft_neuron_config_args = (
        _get_default_neuron_config_for_speculation(
            speculation_config.draft_model_config, parallel_config,
            scheduler_config))
    if is_eagle:
        default_draft_neuron_config_args['is_eagle_draft'] = True
        default_draft_neuron_config_args['has_pre_attention_norm'] = False

    draft_neuron_config = _get_neuron_config_after_override(
        default_draft_neuron_config_args,
        speculation_config.draft_model_config.override_neuron_config)

    draft_model.load_weights(speculation_config.draft_model_config.model,
                             tp_degree=speculation_config.
                             draft_parallel_config.tensor_parallel_size,
                             amp=TORCH_DTYPE_TO_NEURON_AMP[
                                 speculation_config.draft_model_config.dtype],
                             neuron_config=draft_neuron_config,
                             context_length_estimate=context_length_estimates,
                             n_positions=n_positions,
                             batch_size=scheduler_config.max_num_seqs)

    draft_model.eval()

    num_speculative_tokens = speculation_config.num_speculative_tokens
    # Create speculation model instance.
    speculation_model = FusedSpeculativeDecoder(draft_model.model,
                                                target_model.model,
                                                num_speculative_tokens)
    speculation_model.to_neuron()

    return NeuronSpeculationCausalLM(speculation_model)