A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 22:15 UTC
[Misc] Remove dangling references to `--use-v2-block-manager` (#13492)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Feb 19, 2025 at 03:37 UTC
00b69c2d274023c47edc69454573ac656b300828
2 files changed
+2
-3
.buildkite/nightly-benchmarks/tests/serving-tests.json
+1
-2
@@ -66,8 +66,7 @@
66
"swap_space": 16,
67
"speculative_model": "turboderp/Qwama-0.5B-Instruct",
68
"num_speculative_tokens": 4,
69
- "speculative_draft_tensor_parallel_size": 1,
70
- "use_v2_block_manager": ""
69
+ "speculative_draft_tensor_parallel_size": 1
70
},
71
"client_parameters": {
72
"model": "meta-llama/Meta-Llama-3.1-70B-Instruct",
docs/source/features/spec_decode.md
+1
-1
@@ -45,7 +45,7 @@ To perform the same with an online mode launch the server:
45
46
```bash
47
python -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --port 8000 --model facebook/opt-6.7b \
48
- --seed 42 -tp 1 --speculative_model facebook/opt-125m --use-v2-block-manager \
48
+ --seed 42 -tp 1 --speculative_model facebook/opt-125m \
49
--num_speculative_tokens 5 --gpu_memory_utilization 0.8
50
```
51