[doc] Better document for Draft-Target-Model (DTM) speculative decoding #3797
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
wili-65535
added
the
Doc
wili-65535
force-pushed
the
docs/dtm-release-0.19
branch
from
8ecae79 to
25740c4
Compare
achartier
reviewed
Apr 23, 2025
wili-65535
force-pushed
the
docs/dtm-release-0.19
branch
from
05647f2 to
63b42c4
Compare
litaotju
approved these changes
Apr 24, 2025
|
Approving for the change, this is fine to submit to 0.19. @wili-65535 pls let me know to merge it after you addressed all the comments from @achartier . Thx |
Signed-off-by: wili-65535 <[email protected]>
Signed-off-by: wili-65535 <[email protected]>
Signed-off-by: wili-65535 <[email protected]>
wili-65535
force-pushed
the
docs/dtm-release-0.19
branch
from
63b42c4 to
a085fe2
Compare
|
/bot skip --comment "Doc update" |
|
PR_Github #3244 [ skip ] triggered by Bot |
|
PR_Github #3244 [ skip ] completed with state |
|
Thanks a lot @achartier , @litaotju ! |
DomBrown
pushed a commit
to DomBrown/TensorRT-LLM
that referenced
this pull request
Apr 28, 2025
Signed-off-by: ZhanruiSunCh <[email protected]> test: add test cases for 0.19 release (NVIDIA#3608) * fix test name Signed-off-by: Ivy Zhang <[email protected]> * add quickstart test for nemotron-ultra Signed-off-by: Ivy Zhang <[email protected]> * add rcca multi-node test case for deepseek-v3 Signed-off-by: Ivy Zhang <[email protected]> * add rcca info Signed-off-by: Ivy Zhang <[email protected]> --------- Signed-off-by: Ivy Zhang <[email protected]> Signed-off-by: Ivy Zhang <[email protected]> squash (NVIDIA#3642) Signed-off-by: Enwei Zhu <[email protected]> fix: nvbugs/5187237: fix deterministic mode crash (NVIDIA#3448) * nvbugs/5187237 nvbugs/5112075: fix deterministic mode error * remove waive Signed-off-by: Xiwen Yu <[email protected]> * Revert "remove waive" This reverts commit 0bf5486d19906d692bfb7a6262333c296b0087ac. Signed-off-by: Xiwen Yu <[email protected]> * revert ar fusion Signed-off-by: Xiwen Yu <[email protected]> --------- Signed-off-by: Xiwen Yu <[email protected]> update fp8 doc (NVIDIA#3647) Signed-off-by: taoli <[email protected]> Co-authored-by: taoli <[email protected]> tests: change qa perf test to trtllm-bench (NVIDIA#3619) Signed-off-by: Ruodi <[email protected]> Co-authored-by: Larry <[email protected]> fix: FP8 quantized lm_head (NvBug 5214229) (NVIDIA#3567) Signed-off-by: Enwei Zhu <[email protected]> infra: Add PR approval protection for the release branch (NVIDIA#3634) Signed-off-by: Yanchao Lu <[email protected]> fix: nvbugs/5231298: pytorch allreduce issue (NVIDIA#3673) Signed-off-by: Xiwen Yu <[email protected]> Fix: nvbugs/5222698 variable not defined (NVIDIA#3630) * Fix: nvbugs/5222698 variable not defined Signed-off-by: Zongfei Jing <[email protected]> * Tidy code Signed-off-by: Zongfei Jing <[email protected]> --------- Signed-off-by: Zongfei Jing <[email protected]> test:sync waives.txt from main branch by disabling test_perf/gpt_350m-cppmanager case (NVIDIA#3685) Signed-off-by: nv-guomingz <[email protected]> test:restore fp8 kv cache testing for L0 (NVIDIA#3671) Signed-off-by: nv-guomingz <[email protected]> doc: Update DeepSeek perf docs (NVIDIA#3693) * Update DeepSeek perf docs Signed-off-by: Kaiyu Xie <[email protected]> * update Signed-off-by: Kaiyu Xie <[email protected]> * Apply suggestions from code review Co-authored-by: Copilot <[email protected]> Signed-off-by: Kaiyu Xie <[email protected]> --------- Signed-off-by: Kaiyu Xie <[email protected]> Co-authored-by: Copilot <[email protected]> tests: waive test_llm_multi_node (NVIDIA#3664) Signed-off-by: junq <[email protected]> fix: update test_user_buffers_mm_add_prologue atol (NVIDIA#3711) Signed-off-by: Jin Li <[email protected]> Fix: cherry-pick hmac encryption from main branch (NVIDIA#3635) * security fix cherry-pick changes from main Signed-off-by: Yibin Li <[email protected]> * fix hmac in remote mpi session (NVIDIA#3649) Signed-off-by: Yan Chunwei <[email protected]> --------- Signed-off-by: Yibin Li <[email protected]> Signed-off-by: Yan Chunwei <[email protected]> Co-authored-by: Yan Chunwei <[email protected]> Un-waive DS-V3-Lite tests. (NVIDIA#3621) Signed-off-by: Tracin <[email protected]> fix: FP8 kv accuracy (NVIDIA#3675) * fix FP8 kv accuracy Signed-off-by: Dylan Chen <[email protected]> * update doc Signed-off-by: Dylan Chen <[email protected]> --------- Signed-off-by: Dylan Chen <[email protected]> Fix script options for engines. (NVIDIA#3622) Signed-off-by: Tracin <[email protected]> unwaive multi-node test (NVIDIA#3721) Signed-off-by: Superjomn <[email protected]> chore : Split more tests out of gpt tests (NVIDIA#3524) (NVIDIA#3674) Signed-off-by: peaceh <[email protected]> doc:add torch examples link into torch backend documentation (NVIDIA#3749) Signed-off-by: nv-guomingz <[email protected]> Co-authored-by: nv-guomingz <[email protected]> test: Get Eagle tests working (NVIDIA#3593) (NVIDIA#3722) Signed-off-by: Balaram Buddharaju <[email protected]> Co-authored-by: brb-nv <[email protected]> Waive L0 test (NVIDIA#3756) Signed-off-by: Yiqing Yan <[email protected]> waive failed case in perf test, change default max_batch_size to 512 and write config.json to output log (NVIDIA#3656) Signed-off-by: Ruodi <[email protected]> Signed-off-by: Larry <[email protected]> Co-authored-by: Larry <[email protected]> Update ds v3 parameters in stress test. (NVIDIA#3676) waive gemma on L20 (NVIDIA#3766) Signed-off-by: Ivy Zhang <[email protected]> https://nvbugs/5141291: Fix convert.py script for Qwen model. (NVIDIA#3758) Include Qwen2VLDecoderLayer in the smooth_qwen2_model function. Signed-off-by: Yukun He <[email protected]> fix: PP4 fixes and cleanup (NVIDIA#3688) Signed-off-by: Anurag Mukkara <[email protected]> Co-authored-by: Sharan Chetlur <[email protected]> remove benchmark test list (NVIDIA#3643) Signed-off-by: Ivy Zhang <[email protected]> skip disagg deepseek test if sm!=90 (NVIDIA#3720) Signed-off-by: Chuang Zhu <[email protected]> test: skip failed cases on B200 (NVIDIA#3710) * add skip condition to tests Signed-off-by: xinhe-nv <[email protected]> * fix error Signed-off-by: xinhe-nv <[email protected]> --------- Signed-off-by: xinhe-nv <[email protected]> test: [nvbug: 5234494] skip_pre_ada for fp8 cases (NVIDIA#3718) * skip_pre_ada for fp8 cases Signed-off-by: Ivy Zhang <[email protected]> * update Signed-off-by: Ivy Zhang <[email protected]> * update after rebase Signed-off-by: Ivy Zhang <[email protected]> --------- Signed-off-by: Ivy Zhang <[email protected]> add know issue to deepseek doc. (NVIDIA#3800) Signed-off-by: Fanrong Li <[email protected]> Fix ModelOpt Mixtral AWQ OOM (NVIDIA#3714) (NVIDIA#3761) Signed-off-by: Barry Kang <[email protected]> Co-authored-by: Larry <[email protected]> Waive L0 tests (NVIDIA#3826) Signed-off-by: Yiqing Yan <[email protected]> fix: Reduce memory usage in fused moe op associated with AutoTuning and fix moe fallback issue. (NVIDIA#3793) * Reduce memory usage in fused moe op associated with AutoTuning. * Replace pre-defined bucket size strategy with a generating function based on the tune_max_num_tokens. * Add free_memory logic of workspace in min_latency_mode fused moe path. Signed-off-by: Yukun He <[email protected]> * Fix fused_moe fallback issue. (NVIDIA#3652) min_latency_mode is only set to False during warmup phase. Thus when it becomes true during inference, all tactics fall back to the default one and thus cause perf regression. Signed-off-by: Yukun He <[email protected]> --------- Signed-off-by: Yukun He <[email protected]> [doc] Better document for Draft-Target-Model (DTM) speculative decoding (NVIDIA#3797) Signed-off-by: wili-65535 <[email protected]> Signed-off-by: Dom Brown <[email protected]> Fix pre-commit Signed-off-by: Dom Brown <[email protected]> Fix again Signed-off-by: Dom Brown <[email protected]> Address some review comments for the MI Signed-off-by: Dom Brown <[email protected]>
dcampora
pushed a commit
that referenced
this pull request
Apr 29, 2025
test: add test cases for 0.19 release (#3608) * fix test name * add quickstart test for nemotron-ultra * add rcca multi-node test case for deepseek-v3 * add rcca info --------- squash (#3642) fix: nvbugs/5187237: fix deterministic mode crash (#3448) * nvbugs/5187237 nvbugs/5112075: fix deterministic mode error * remove waive * Revert "remove waive" This reverts commit 0bf5486d19906d692bfb7a6262333c296b0087ac. * revert ar fusion --------- update fp8 doc (#3647) tests: change qa perf test to trtllm-bench (#3619) fix: FP8 quantized lm_head (NvBug 5214229) (#3567) infra: Add PR approval protection for the release branch (#3634) fix: nvbugs/5231298: pytorch allreduce issue (#3673) Fix: nvbugs/5222698 variable not defined (#3630) * Fix: nvbugs/5222698 variable not defined * Tidy code --------- test:sync waives.txt from main branch by disabling test_perf/gpt_350m-cppmanager case (#3685) test:restore fp8 kv cache testing for L0 (#3671) doc: Update DeepSeek perf docs (#3693) * Update DeepSeek perf docs * update * Apply suggestions from code review --------- tests: waive test_llm_multi_node (#3664) fix: update test_user_buffers_mm_add_prologue atol (#3711) Fix: cherry-pick hmac encryption from main branch (#3635) * security fix cherry-pick changes from main * fix hmac in remote mpi session (#3649) --------- Un-waive DS-V3-Lite tests. (#3621) fix: FP8 kv accuracy (#3675) * fix FP8 kv accuracy * update doc --------- Fix script options for engines. (#3622) unwaive multi-node test (#3721) chore : Split more tests out of gpt tests (#3524) (#3674) doc:add torch examples link into torch backend documentation (#3749) test: Get Eagle tests working (#3593) (#3722) Waive L0 test (#3756) waive failed case in perf test, change default max_batch_size to 512 and write config.json to output log (#3656) Update ds v3 parameters in stress test. (#3676) waive gemma on L20 (#3766) https://nvbugs/5141291: Fix convert.py script for Qwen model. (#3758) Include Qwen2VLDecoderLayer in the smooth_qwen2_model function. fix: PP4 fixes and cleanup (#3688) remove benchmark test list (#3643) skip disagg deepseek test if sm!=90 (#3720) test: skip failed cases on B200 (#3710) * add skip condition to tests * fix error --------- test: [nvbug: 5234494] skip_pre_ada for fp8 cases (#3718) * skip_pre_ada for fp8 cases * update * update after rebase --------- add know issue to deepseek doc. (#3800) Fix ModelOpt Mixtral AWQ OOM (#3714) (#3761) Waive L0 tests (#3826) fix: Reduce memory usage in fused moe op associated with AutoTuning and fix moe fallback issue. (#3793) * Reduce memory usage in fused moe op associated with AutoTuning. * Replace pre-defined bucket size strategy with a generating function based on the tune_max_num_tokens. * Add free_memory logic of workspace in min_latency_mode fused moe path. * Fix fused_moe fallback issue. (#3652) min_latency_mode is only set to False during warmup phase. Thus when it becomes true during inference, all tactics fall back to the default one and thus cause perf regression. --------- [doc] Better document for Draft-Target-Model (DTM) speculative decoding (#3797) Fix pre-commit Fix again Address some review comments for the MI Signed-off-by: Dom Brown <[email protected]> Co-authored-by: Zhanrui Sun <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Previous PR3366 about DTM document, cherry-pick from PR3655.
A more detailed document about using Draft-Target-Model in Triton Inference Server.
Target of this PR: