Commit 43002c5

committed

Revert "[TRTLLM-11540][feat] Add EAGLE3 dynamic tree speculative decoding support (NVIDIA#12062)"

This reverts commit 4ece13c. Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>

1 parent 11716e0 commit 43002c5Copy full SHA for 43002c5

36 files changed

+414

-3566

lines changed

cpp/tensorrt_llm
- kernels
  - speculativeDecoding
  - trtllmGenKernels/fmha
    - fmhaRunnerParams.h
    - prepareCustomMask.cu
  - xqaDispatcher.cpp
- thop
docs/source
- features
  - feature-combination-matrix.md
  - speculative-decoding.md
- models
  - supported-models.md
examples/llm-api
- quickstart_advanced.py
tensorrt_llm
- _torch
  - attention_backend
    - interface.py
    - sparse
      - dsa.py
    - trtllm.py
  - modules
    - attention.py
  - pyexecutor
  - speculative
- llmapi
  - llm_args.py
tests/unittest
- _torch
  - modeling
    - test_modeling_llama.py
  - speculative
  - thop/parallel
    - test_custom_ops.py
- others
  - test_kv_cache_update.py

36 files changed

+414

-3566

lines changed

`‎cpp/tensorrt_llm/kernels/speculativeDecoding/dynamicTreeKernels.cu‎`

This file was deleted.

Comments

(0)