This pipeline provides optimized support for the DeepSeek Coder V1.5 code completion model, a member of the LlamaForCausalLM architecture family. DeepSeek Coder is an open source code generation model released by DeepSeek.
Read the code and instructions, and run on desktop.
This pipeline provides optimized support for the DeepSeek Coder V1.5 code completion model, a member of the LlamaForCausalLM architecture family. DeepSeek Coder is an open source code generation model released by DeepSeek.
Read the code and instructions, and run on desktop.
Language: Python
API: MAX Graph
This pipeline provides optimized support for the DeepSeek Coder V1.5 code
completion model, a member of the LlamaForCausalLM architecture family.
The model itself has been constructed in Python
using the MAX Graph API.
The MAX Graph API provides an accessible interface to the construction of flexible accelerated compute graphs, which are then optimized by the MAX Engine's advanced graph compiler. This pipeline showcases how a large language model can be fully defined using Python and MAX Graphs and then compiled for optimal inference performance via the MAX Engine.
DeepSeek Coder is an open
source code generation model released by DeepSeek. The core architecture of the
V1.5 version of the model is based on the LlamaForCausalLM base large
language model architecture.
The easiest way to try out this pipeline is with our Magic command-line tool.
Install Magic on macOS and Ubuntu with this command:
curl -ssL https://magic.modular.com | bash
Then run the source command that's printed in your terminal.
To see the available commands, you can run magic --help.
Learn more about Magic here.
Clone the MAX examples repository:
If you don't already have a local clone of this repository, create one via:
git clone https://github.com/modularml/max.git
The following instructions assume that you're present within this pipeline's directory, and you can change to it after cloning:
cd max/pipelines/python/
Host a code generation completion endpoint via MAX Serve.
MAX Serve provides functionality to host performant OpenAI compatible endpoints using the FastAPI framework.
You can configure the pipeline to be hosted by using the serve command.
For example:
magic run serve --huggingface-repo-id deepseek-ai/deepseek-coder-7b-instruct-v1.5
A request can be submitted via a cURL command.
curl -N http://localhost:8000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-ai/deepseek-coder-7b-instruct-v1.5",
"stream": true,
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Please complete the following Python function: `def fibonacci(n):`"}
]
}'
The following command-line options are available to customize operation of the pipeline:
--huggingface-repo-id: Specify the repository ID of a Hugging Face model
repository to use. This is used to load tokenizers, architectures and model
weights.--force-download: Specify whether to force a download of configuration
files and weights even if they already exist in the local cache. Set this
if you want to ensure you have the correct version of the model.--max-cache-batch-size: Specifies the maximum batch size to be used.
Default is 1.--max-ce-batch-size: Set the maximum cache size reserved for a single
context encoding batch. The effective limit will be the lesser of this value
and max-cache-batch-size.
Default is 32.--max-length: Controls the maximum length of the text sequence
(includes the input tokens).
(Default value: 512)--max-new-tokens: The maximum number of new tokens to generate. If a -1
value is provided, the model will continue to generate tokens for the entire
context length. (Default value: -1)--quantization-encoding: The encoding to use for a datatype that can be
quantized to a low bits per weight format.
Valid values: q4_0, q4_k, q6_k, bfloat16, float32.
(Default value: bfloat16).--save-to-serialized-model-path: If specified, writes the serialized model
to this path.--serialized-model-path: If specified, tries to load a serialized model
from this path.--top-k: Limits the sampling to the K most probable tokens. Default is 1.--trust-remote-code: Indicate whether to allow custom modeling files from
Hugging Face repositories. Set this to true with caution, as it may
introduce security risks.--use-gpu: Uses the GPU to execute the model. A device ID can optionally
be provided to execute on a specific GPU in the system.--weight-path: Overrides the default URL, and allows for an
already-downloaded pretrained weight file to be used with the model.279 lines - 10.8 KB
Graph
# ===----------------------------------------------------------------------=== #
# Copyright (c) 2024, Modular Inc. All rights reserved.
#
# Licensed under the Apache License v2.0 with LLVM Exceptions:
# https://llvm.org/LICENSE.txt
#
# 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.
# ===----------------------------------------------------------------------=== #
from __future__ import annotations
import logging
import numpy as np
from dataprocessing import batch_padded_tokens_and_mask
from max.driver import Tensor
from max.dtype import DType
from max.engine import InferenceSession, Model
from max.pipelines import ModelOutputs, PipelineModel, TextContext
from max.pipelines.kv_cache import (
KVCacheManager,
KVCacheParams,
KVCacheStrategy,
estimate_kv_cache_size,
load_kv_manager,
)
from .graph import _build_graph
class CoderModel(PipelineModel):
def execute(self, *model_inputs: Tensor) -> ModelOutputs:
model_outputs = self.model.execute(
*model_inputs,
copy_inputs_to_device=(
self.pipeline_config.cache_strategy == KVCacheStrategy.NAIVE
),
)
if self.pipeline_config.enable_echo:
return ModelOutputs(
next_token_logits=model_outputs[0], logits=model_outputs[1]
)
else:
return ModelOutputs(next_token_logits=model_outputs[0])
def _prepare_continuous_initial_token_inputs(
self, context_batch: list[TextContext]
) -> tuple[Tensor, ...]:
# Get tokens and seq_ids
tokens = [ctx.next_tokens for ctx in context_batch]
# Get input_row_offsets: start and end position of each batch in the
# combined total_seq_len dimension.
input_row_offsets = Tensor.from_numpy(
np.cumsum(
[0] + [ctx.seq_len for ctx in context_batch],
dtype=np.uint32,
)
).to(self.pipeline_config.device)
# Create a ragged token vector of length: sum(len(t) for t in tokens).
next_tokens_batch = np.concatenate(tokens)
next_tokens_batch = Tensor.from_numpy(next_tokens_batch).to(
self.pipeline_config.device
)
return (next_tokens_batch, input_row_offsets)
def _prepare_naive_initial_token_inputs(
self, context_batch: list[TextContext]
) -> tuple[Tensor, ...]:
# Get tokens and seq_ids
tokens = [ctx.next_tokens for ctx in context_batch]
# Pad tokens and compute attention mask for the batch.
max_seq_len = self.kv_manager.max_sequence_length
start_pos = [max_seq_len] * len(context_batch)
next_tokens_batch, _, attn_mask = batch_padded_tokens_and_mask(
start_pos=start_pos,
tokens=tokens,
pad_to_multiple_of=self.pipeline_config.pad_to_multiple_of,
)
return (next_tokens_batch, attn_mask)
def prepare_initial_token_inputs(
self, context_batch: list[TextContext]
) -> tuple[Tensor, ...]:
"""Prepare the inputs for the first pass in multistep execution."""
if self.pipeline_config.cache_strategy == KVCacheStrategy.CONTINUOUS:
return self._prepare_continuous_initial_token_inputs(context_batch)
else:
return self._prepare_naive_initial_token_inputs(context_batch)
def _prepare_continuous_next_token_inputs(
self,
next_tokens: Tensor,
prev_model_inputs: tuple[Tensor, ...],
):
_, old_row_offsets = prev_model_inputs
row_offsets_size = old_row_offsets.shape[0]
next_row_offsets = self._input_row_offsets_prealloc[:row_offsets_size]
next_token_inputs = (next_tokens, next_row_offsets)
return next_token_inputs
def _prepare_naive_next_token_inputs(
self,
next_tokens: Tensor,
prev_model_inputs: tuple[Tensor, ...],
):
prev_tokens, prev_attn_mask = prev_model_inputs
batch_size = prev_tokens.shape[0]
start_pos = [prev_attn_mask.shape[-1]] * batch_size
next_tokens_batch, _, attn_mask = batch_padded_tokens_and_mask(
start_pos=start_pos,
tokens=next_tokens,
pad_to_multiple_of=self.pipeline_config.pad_to_multiple_of,
)
next_token_inputs = (next_tokens_batch, attn_mask)
return next_token_inputs
def prepare_next_token_inputs(
self,
next_tokens: Tensor,
prev_model_inputs: tuple[Tensor, ...],
) -> tuple[Tensor, ...]:
"""Prepare the inputs for the next token in multistep execution.
This should avoid any device synchronization or copy operations.
"""
if self.pipeline_config.cache_strategy == KVCacheStrategy.CONTINUOUS:
return self._prepare_continuous_next_token_inputs(
next_tokens, prev_model_inputs
)
else:
return self._prepare_naive_next_token_inputs(
next_tokens, prev_model_inputs
)
def _get_kv_params(self) -> KVCacheParams:
cache_dtype = (
DType.float32
if self.pipeline_config.quantization_encoding.quantization_encoding
is not None
else self.pipeline_config.dtype
)
return KVCacheParams(
dtype=cache_dtype,
n_kv_heads=self.pipeline_config.huggingface_config.num_key_value_heads,
head_dim=self.pipeline_config.huggingface_config.hidden_size
// self.pipeline_config.huggingface_config.num_attention_heads,
cache_strategy=self.pipeline_config.cache_strategy,
)
def load_kv_manager(self, session: InferenceSession) -> KVCacheManager:
return load_kv_manager(
params=self._get_kv_params(),
max_cache_batch_size=self.pipeline_config.max_cache_batch_size,
max_seq_len=self.pipeline_config.huggingface_config.max_seq_len,
num_layers=self.pipeline_config.huggingface_config.num_hidden_layers,
devices=[self.pipeline_config.device],
session=session,
)
def estimate_kv_cache_size(self) -> int:
return estimate_kv_cache_size(
params=self._get_kv_params(),
max_cache_batch_size=self.pipeline_config.max_cache_batch_size,
max_seq_len=self.pipeline_config.huggingface_config.max_seq_len,
num_layers=self.pipeline_config.huggingface_config.num_hidden_layers,
devices=[self.pipeline_config.device],
)
def load_model(
self,
session: InferenceSession,
) -> Model:
# Pre-allocate a buffer for input_row_offsets in multistep execution.
# We do this to avoid materializing and copying a buffer with each multistep step
self._input_row_offsets_prealloc = Tensor.from_numpy(
np.arange(
self.pipeline_config.max_cache_batch_size + 1, dtype=np.uint32
)
).to(self.pipeline_config.device)
# Read in weights.
self._weights = self.pipeline_config.load_weights()
if serialized_path := self.pipeline_config.serialized_model_path:
# Hydrate all weights to be referenced by the serialized path.
weights_registry = {}
for name, tensor in self._weights._tensors.items():
weights_registry[name] = tensor.data
logging.info("Loading serialized model from ", serialized_path)
return session.load(
serialized_path, weights_registry=weights_registry
)
else:
logging.info("Building model...")
graph = _build_graph(
self.pipeline_config,
self._weights,
self._get_kv_params(),
kv_manager=self.kv_manager,
)
logging.info("Compiling...")
model = session.load(
graph,
weights_registry=self._weights.allocated_weights, # type: ignore
)
if (
export_path
:= self.pipeline_config.save_to_serialized_model_path
):
logging.info("Exporting serialized model to %s", export_path)
model._export_mef(export_path)
return model
def compute_log_probabilities(
self,
model_inputs: Sequence[Tensor],
model_outputs: ModelOutputs,
next_tokens: Tensor,
batch_top_n: list[int],
batch_echo: list[bool],
) -> list[LogProbabilities | None] | None:
if any(echo for echo in batch_echo):
if model_outputs.logits is None:
warnings.warn(
"Could not get logprobs with echo because the full logits"
f" were not returned by {self.pipeline_config.short_name}"
" model. Please ensure that this model is started with "
"`--enable-echo`."
)
assert (
not self.pipeline_config.enable_echo
), "Echo was enabled but logits were not returned."
return None
logits = model_outputs.logits.to(CPU()).to_numpy()
next_token_logits = model_outputs.next_token_logits.to(CPU()).to_numpy()
sampled_tokens = next_tokens.to(CPU()).to_numpy()
# Handle batched inputs.
token_tensor, _, valid_length_tensor = model_inputs
tokens = token_tensor.to(CPU()).to_numpy()
valid_lengths = valid_length_tensor.to(CPU()).to_numpy()
def _get_logits_and_samples(
batch_index: int, echo: bool
) -> tuple[np.ndarray, np.ndarray]:
if echo:
seq_len = valid_lengths[batch_index]
padded_tokens = tokens[batch_index]
assert model_outputs.logits is not None
batch_logits = logits[batch_index, :seq_len]
samples = np.concatenate(
(
padded_tokens[1:seq_len],
sampled_tokens[batch_index : batch_index + 1],
)
)
else:
batch_logits = next_token_logits[batch_index : batch_index + 1]
samples = sampled_tokens[batch_index : batch_index + 1]
return batch_logits, samples
return compute_log_probabilities(
_get_logits_and_samples, batch_top_n, batch_echo
)
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
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.
---- LLVM Exceptions to the Apache 2.0 License ----
As an exception, if, as a result of your compiling your source code, portions of this Software are embedded into an Object form of such source code, you may redistribute such embedded portions in such Object form without complying with the conditions of Sections 4(a), 4(b) and 4(d) of the License.
In addition, if you combine or link compiled forms of this Software with software that is licensed under the GPLv2 ("Combined Software") and if a court of competent jurisdiction determines that the patent provision (Section 3), the indemnity provision (Section 9) or other Section of the License conflicts with the conditions of the GPLv2, you may retroactively and prospectively choose to deem waived or otherwise exclude such Section(s) of the License, but only in their entirety and only with respect to the Combined Software.
The LLVM Project contains third party software which is under different license terms. All such code will be identified clearly using at least one of two mechanisms:
LICENSE.txt or
LICENSE file at the top containing the specific license and restrictions
which apply to that software, or@ Copyright - Modular Inc - 2024