llama.cpp/examples/parallel/README.md

# llama.cpp/example/parallel

Simplified simulation of serving incoming requests in parallel

## Example

Generate 128 client requests (`-ns 128`), simulating 8 concurrent clients (`-np 8`). The system prompt is shared (`-pps`), meaning that it is computed once at the start. The client requests consist of up to 10 junk questions (`--junk 10`) followed by the actual question.

```bash
llama-parallel -m model.gguf -np 8 -ns 128 --top-k 1 -pps --junk 10 -c 16384
```

> [!NOTE]
> It's recommended to use base models with this example. Instruction tuned models might not be able to properly follow the custom chat template specified here, so the results might not be as expected.
llama : custom attention mask + parallel decoding + no context swaps (#3228) * tests : verify that RoPE is "additive" * llama : replace ggml_diag_mask_inf with ggml_add (custom -inf mask) * ggml : ggml_rope now takes a vector with positions instead of n_past * metal : add rope_f16 kernel + optimize cpy kernels * llama : unified KV cache + batch inference API * llama : add new llama_decode() API that works with llama_batch * llama : add cell_max heuristic for more efficient kv_cache * llama : extend llama_kv_cache API * llama : more robust cell_max heuristic + wip shift * metal : disable concurrency optimization * llama : add llama_kv_cache_shift_seq + no more context swaps * llama : apply K-cache roping for Falcon and Baichuan * speculative : fix KV cache management * parallel : example for serving multiple users in parallel * parallel : disable hot-plug to avoid cache fragmentation * fixes : speculative KV cache + llama worst-case graph * llama : extend batch API to select which logits to output * llama : fix worst case graph build * ggml-cuda : update rope implementation for parallel decoding (#3254) * ggml-cuda : update rope implementation for parallel decoding * better solution for p0 computation * fix rope * simpler rope implementation --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * make : add parallel to build + fix static functions in llama.cpp * simple : fix token counting * parallel : various improvements * llama : fix cell_max logic + rename functions * parallel : try smaller batches when the KV cache is fragmented * parallel : fix sequence termination criteria * llama : silence errors KV cache errors * parallel : remove new line from prompt * parallel : process system prompt once + configurable paramters + llama API * parallel : remove question with short answers * parallel : count cache misses * parallel : print misses on each request * parallel : minor * llama : fix n_kv to never become 0 * parallel : rename hot-plug to continuous-batching * llama : improve llama_batch API + simplify parallel example * simple : add parallel decoding support * simple : improve comments + free batch * ggml-cuda : add rope f16, restore performance with parallel decoding (#3272) * ggml-cuda : add rope f16, restore performance * offload KQ_mask with all models * fix rope shift --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * llama : disable MPI for now ggml-ci * train : make KQ_pos memory buffer permanent via dummy scale op * ggml : revert change to ggml_cpy, add ggml_cont_Nd instead (#3275) ggml-ci * parallel : fix bug (extra BOS) + smaller token_prev array * parallel : fix cases where the input prompts can overflow the batch * parallel : add disabled experimental batch chunking in powers of two * llama : llama.h formatting + comments * simple : add README.md * llama : fix kv cache heuristic when context is less than 32 * parallel : fix crash when `-n -1` * llama : simplify returns if/else branches * metal : use mm kernels for batch size > 2 * examples : utilize new llama_get_logits_ith() * examples : add example for batched decoding * examples : do not eval prompt 2 times (close #3348) * server : clear the KV cache beyond n_past before llama_decode * server : avoid context swaps by shifting the KV cache --------- Co-authored-by: slaren <slarengh@gmail.com> 2023-09-28 19:04:36 +03:00			`# llama.cpp/example/parallel`

Fix some documentation typos/grammar mistakes (#4032) * typos * Update examples/parallel/README.md Co-authored-by: Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com> --------- Co-authored-by: Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com> 2023-11-11 22:04:58 -08:00			`Simplified simulation of serving incoming requests in parallel`
parallel : add option for non-shared and larger prompts (#13598) * parallel : add option for non-shared and larger prompts * parallel : update readme [no ci] * cont : add note about base models [no ci] * parallel : better var name ggml-ci 2025-05-17 12:58:55 +03:00
			`## Example`

parallel : increase the variability of the prompt lengths (#13927) ggml-ci 2025-05-30 19:38:07 +03:00			Generate 128 client requests (`-ns 128`), simulating 8 concurrent clients (`-np 8`). The system prompt is shared (`-pps`), meaning that it is computed once at the start. The client requests consist of up to 10 junk questions (`--junk 10`) followed by the actual question.
parallel : add option for non-shared and larger prompts (#13598) * parallel : add option for non-shared and larger prompts * parallel : update readme [no ci] * cont : add note about base models [no ci] * parallel : better var name ggml-ci 2025-05-17 12:58:55 +03:00
			```bash
			`llama-parallel -m model.gguf -np 8 -ns 128 --top-k 1 -pps --junk 10 -c 16384`
			```

			`> [!NOTE]`
			`> It's recommended to use base models with this example. Instruction tuned models might not be able to properly follow the custom chat template specified here, so the results might not be as expected.`