phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-06-28 02:53:30 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

refactor: move all llama code to models/llama out of meta reference (#1887)

Browse source 
# What does this PR do?

Move around bits. This makes the copies from llama-models _much_ easier
to maintain and ensures we don't entangle meta-reference specific
tidbits into llama-models code even by accident.

Also, kills the meta-reference-quantized-gpu distro and rolls
quantization deps into meta-reference-gpu.

## Test Plan

```
LLAMA_MODELS_DEBUG=1 \
  with-proxy llama stack run meta-reference-gpu \
  --env INFERENCE_MODEL=meta-llama/Llama-4-Scout-17B-16E-Instruct \
   --env INFERENCE_CHECKPOINT_DIR=<DIR> \
   --env MODEL_PARALLEL_SIZE=4 \
   --env QUANTIZATION_TYPE=fp8_mixed
```

Start a server with and without quantization. Point integration tests to
it using:

```
pytest -s -v  tests/integration/inference/test_text_inference.py \
   --stack-config http://localhost:8321 --text-model meta-llama/Llama-4-Scout-17B-16E-Instruct
```
This commit is contained in:
Ashwin Bharambe 2025-04-07 15:03:58 -07:00 • committed by GitHub
parent c52ccc4bbd
commit 530d4bdfe1
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
85 changed files with 1267 additions and 1683 deletions
Download patch file Download diff file Expand all files Collapse all files

25
llama_stack/models/llama/llama4/tokenizer.py
View file

@ -4,9 +4,6 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
# Copyright (c) Meta Platforms, Inc. and affiliates.
# This software may be used and distributed in accordance with the terms of the Llama 3 Community License Agreement.
import os
from logging import getLogger
from pathlib import Path
@ -59,8 +56,6 @@ LLAMA4_TEXT_POST_TRAIN_SPECIAL_TOKENS = [
"<|text_post_train_reserved_special_token_3|>",
"<|text_post_train_reserved_special_token_4|>",
"<|text_post_train_reserved_special_token_5|>",
"<|python_start|>",
"<|python_end|>",
"<|finetune_right_pad|>",
] + get_reserved_special_tokens(
"text_post_train", 61, 6
@ -85,8 +80,23 @@ LLAMA4_VISION_SPECIAL_TOKENS = [
"vision", 1041, 7
) # <|vision_reserved_special_token_7|>, ..., <|vision_reserved_special_token_1047|>
# 201134, ..., 201143
LLAMA4_REASONING_SPECIAL_TOKENS = [
"<|reasoning_reserved_special_token_0|>",
"<|reasoning_reserved_special_token_1|>",
"<|reasoning_reserved_special_token_2|>",
"<|reasoning_reserved_special_token_3|>",
"<|reasoning_reserved_special_token_4|>",
"<|reasoning_reserved_special_token_5|>",
"<|reasoning_reserved_special_token_6|>",
"<|reasoning_reserved_special_token_7|>",
"<|reasoning_thinking_start|>",
"<|reasoning_thinking_end|>",
]
LLAMA4_SPECIAL_TOKENS = LLAMA4_TEXT_POST_TRAIN_SPECIAL_TOKENS + LLAMA4_VISION_SPECIAL_TOKENS
LLAMA4_SPECIAL_TOKENS = (
LLAMA4_TEXT_POST_TRAIN_SPECIAL_TOKENS + LLAMA4_VISION_SPECIAL_TOKENS + LLAMA4_REASONING_SPECIAL_TOKENS
)
BASIC_SPECIAL_TOKENS = [
"<|begin_of_text|>",
@ -155,6 +165,9 @@ class Tokenizer:
self.eot_id: int = self.special_tokens["<|eot|>"]
self.eom_id: int = self.special_tokens["<|eom|>"]
self.thinking_start_id: int = self.special_tokens["<|reasoning_thinking_start|>"]
self.thinking_end_id: int = self.special_tokens["<|reasoning_thinking_end|>"]
self.stop_tokens = [
self.eos_id,
self.special_tokens["<|eom|>"],