Open Source LLM Inference Engines

Overview of popular open source large language model inference engines. An inference engine is the program which loads a models weights and generates text responses based on given inputs.

• View feature table
• View benchmarks

Feel free to create a PR or issue if you want a new engine column, feature row, or update a status.

Compared Inference Engines

• vLLM: Designed to provide SOTA throughput.
• TensorRT-LLM: Nvidias design for a high performance extensible pytorch-like API for use with Nvidia Triton Inference Server.
• llama.cpp: Pure C++ without any dependencies, with Apple Silicon prioritized.
• TGI: HuggingFace' fast and flexible engine designed for high throughput.
• LightLLM: Lightweight, fast and flexible framework targeting performance, written purely in Python / Triton.
• DeepSpeed-MII / DeepSpeed-FastGen: Microsofts high performance implementation including SOTA Dynamic Splitfuse
• ExLlamaV2: Efficiently run language models on modern consumer GPUs. Implements SOTA quantization method, EXL2.

Comparison Table

✅ Included | 🟠 Inferior Alternative | 🌩️ Exists but has Issues | 🔨 PR | 🗓️ Planned |❓ Unclear / Unofficial | ❌ Not Implemented

	vLLM	TensorRT-LLM	llama.cpp	TGI	LightLLM	Fastgen	ExLlamaV2
Optimizations
FlashAttention2	✅ 1	✅ 2	🟠 3	✅ 4	✅	✅	✅
PagedAttention	✅ 4	✅ 2	❌ 5	✅	🟠*** 6	✅	✅ 7
Speculative Decoding	🔨 8	✅ 9	✅ 10	✅ 11	❌	❌ 12	✅
Tensor Parallel	✅	✅ 13	🟠** 14	✅ 15	✅	✅ 16	❌
Pipeline Parallel	✅ 17	✅ 18	❌ 19	❓ 15	❌	❌ 20	❌
Optim. / Scheduler
Dyn. SplitFuse (SOTA21)	🗓️ 21	🗓️ 22	❌	❌	❌	✅ 21	❌
Efficient Rtr (better)	❌	❌	❌	❌	✅ 23	❌	❌
Cont. Batching	✅ 21	✅ 24	✅	✅	❌	✅ 16	❓ 25
Optim. / Quant
EXL2 (SOTA26)	🔨 27	❌	❌	✅ 28	❌	❌	✅
AWQ	🌩️ 29	✅	❌	✅	❌	❌	❌
Other Quants	(yes) 30	GPTQ	GGUF 31	(yes) 32	?	?	?
Features
OpenAI-Style API	✅	❌ 33	✅ [^13]	✅ 34	✅ 35	❌	❌
Feat. / Sampling
Beam Search	✅	✅ 2	✅ 36	🟠**** 37	❌	❌ 38	❌ 39
JSON / Grammars via Outlines	✅	🗓️	✅	✅	?	?	✅
Models
Llama 2 / 3	✅	✅	✅	✅	✅	✅	✅
Mistral	✅	✅	✅	✅	✅ 40	✅	✅
Mixtral	✅	✅	✅	✅	✅	✅	✅
Implementation
Core Language	Python	C++	C++	Py / Rust	Python	Python	Python
GPU API	CUDA*	CUDA*	Metal / CUDA	CUDA*	Triton / CUDA	CUDA*	CUDA
Repo
License	Apache 2	Apache 2	MIT	Apache 2 41	Apache 2	Apache 2	MIT
Github Stars	17K	6K	54K	8K	2K	2K	3K

Benchmarks

• BentoML (June 5th, 2024): Compares LMDeploy, MLC-LLM, TGI, TRT-LLM, vLLM

Notes

*Supports Triton for one-off such as FlashAttention (FusedAttention) / quantization, or allows Triton plugins, however the project doesn't use Triton otherwise.

**Sequentially processed tensor split

***"TokenAttention is the special case of PagedAttention when block size equals to 1, which we have tested before and find it under-utilizes GPU compute compared to larger block size. Unless LightLLM's Triton kernel implementation is surprisingly fast, this should not bring speedup."

****TGI maintainers suggest using best_of instead of beam search. (best_of creates n generations and selects the one with the lowest logprob). Anecdotally, beam search is much better at finding the best generation for "non-creative" tasks.

Footnotes

1. https://github.com/vllm-project/vllm/issues/485#issuecomment-1693009046 ↩

2. https://github.com/NVIDIA/TensorRT-LLM/blob/main/docs/source/gpt_attention.md ↩ ↩² ↩³

3. https://github.com/ggerganov/llama.cpp/pull/5021 FlashAttention, but not FlashAttention2 ↩

4. https://github.com/huggingface/text-generation-inference/issues/753#issuecomment-1663525606 ↩ ↩²

5. https://github.com/ggerganov/llama.cpp/issues/1955 ↩

6. https://github.com/ModelTC/lightllm/blob/main/docs/TokenAttention.md ↩

7. https://github.com/turboderp/exllamav2/commit/affc3508c1d18e4294a5062f794f44112a8b07c5 ↩

8. https://github.com/vllm-project/vllm/pull/1797 ↩

9. https://nvidia.github.io/TensorRT-LLM/advanced/speculative-decoding.html ↩

10. https://github.com/ggerganov/llama.cpp/blob/fe680e3d1080a765e5d3150ffd7bab189742898d/examples/speculative/README.md ↩

11. https://github.com/huggingface/text-generation-inference/pull/1308 ↩

12. https://github.com/microsoft/DeepSpeed-MII/issues/254 ↩

13. https://github.com/NVIDIA/TensorRT-LLM/blob/main/cpp/tensorrt_llm/pybind/bindings.cpp#L184 ↩

14. https://github.com/ggerganov/llama.cpp/issues/4014#issuecomment-1804925896 ↩

15. https://github.com/huggingface/text-generation-inference/issues/1031#issuecomment-1727976990 ↩ ↩²

16. https://github.com/microsoft/DeepSpeed-MII ↩ ↩²

17. https://github.com/vllm-project/vllm/issues/387 ↩

18. https://github.com/NVIDIA/TensorRT-LLM/blob/bf0a5afc92f4b2b3191e9e55073953c1f600cf2d/tensorrt_llm/auto_parallel/config.py#L35 ↩

19. "without specific architecture tricks, you will only be using one GPU at a time, and your performance will suffer compared to a single GPU due to communication and synchronization overhead." https://github.com/ggerganov/llama.cpp/issues/4238#issuecomment-1832768597 ↩

20. https://github.com/microsoft/DeepSpeed-MII/issues/329#issuecomment-1830317364 ↩

21. https://blog.vllm.ai/2023/11/14/notes-vllm-vs-deepspeed.html, https://github.com/vllm-project/vllm/issues/1562 ↩ ↩² ↩³ ↩⁴

22. https://github.com/NVIDIA/TensorRT-LLM/issues/317#issuecomment-1810841752 ↩

23. https://github.com/ModelTC/lightllm/blob/a9cf0152ad84beb663cddaf93a784092a47d1515/docs/LightLLM.md#efficient-router ↩

24. https://github.com/NVIDIA/TensorRT-LLM/blob/main/README.md ↩

25. https://github.com/turboderp/exllamav2/discussions/19#discussioncomment-6989460 ↩

26. https://oobabooga.github.io/blog/posts/gptq-awq-exl2-llamacpp/https://oobabooga.github.io/blog/posts/gptq-awq-exl2-llamacpp/#pareto-frontiers ↩

27. https://github.com/vllm-project/vllm/issues/296 ↩

28. https://github.com/huggingface/text-generation-inference/pull/1211 ↩

29. https://github.com/vllm-project/vllm/blob/main/docs/source/quantization/auto_awq.rst ↩

30. https://github.com/vllm-project/vllm/blob/1f24755bf802a2061bd46f3dd1191b7898f13f45/vllm/model_executor/quantization_utils/squeezellm.py#L8 ↩

31. https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/README.md ↩

32. https://github.com/huggingface/text-generation-inference/blob/main/server/text_generation_server/cli.py#L15-L21 ↩

33. https://github.com/NVIDIA/TensorRT-LLM/issues/334 ↩

34. https://huggingface.co/docs/text-generation-inference/messages_api ↩

35. https://github.com/ModelTC/lightllm/blob/main/lightllm/server/api_models.py#L9 ↩

36. https://github.com/ggerganov/llama.cpp/tree/master/examples/beam-search ↩

37. https://github.com/huggingface/text-generation-inference/issues/722#issuecomment-1658823644 ↩

38. https://github.com/microsoft/DeepSpeed-MII/issues/286#issuecomment-1808510043 ↩

39. https://github.com/turboderp/exllamav2/issues/84 ↩

40. https://github.com/ModelTC/lightllm/issues/224#issuecomment-1827365514 ↩

41. https://raw.githubusercontent.com/huggingface/text-generation-inference/main/LICENSE, https://twitter.com/julien_c/status/1777328456709062848 ↩