mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-11-01 07:30:17 +01:00
f30ea47a87
* llama : add pipeline parallelism support for batch processing with multiple CUDA GPUs ggml-ci * server : add -ub, --ubatch-size parameter * fix server embedding test * llama : fix Mamba inference for pipeline parallelism Tested to work correctly with both `main` and `parallel` examples. * llama : limit max batch size to n_batch * add LLAMA_SCHED_MAX_COPIES to configure the number of input copies for pipeline parallelism default increase to 4 (from 2) changing this value may improve performance for some systems, but increases memory usage * fix hip build * fix sycl build (disable cpy_tensor_async) * fix hip build * llama : limit n_batch and n_ubatch to n_ctx during context creation * llama : fix norm backend * batched-bench : sync after decode * swiftui : sync after decode * ggml : allow ggml_get_rows to use multiple threads if they are available * check n_ubatch >= n_tokens with non-casual attention * llama : do not limit n_batch to n_ctx with non-casual attn * server : construct batch with size of llama_n_batch * ggml_backend_cpu_graph_compute : fix return value when alloc fails * llama : better n_batch and n_ubatch comment * fix merge * small fix * reduce default n_batch to 2048 --------- Co-authored-by: Francis Couture-Harpin <git@compilade.net> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
96 lines
2.4 KiB
Gherkin
96 lines
2.4 KiB
Gherkin
@llama.cpp
|
|
@embeddings
|
|
Feature: llama.cpp server
|
|
|
|
Background: Server startup
|
|
Given a server listening on localhost:8080
|
|
And a model file bert-bge-small/ggml-model-f16.gguf from HF repo ggml-org/models
|
|
And a model alias bert-bge-small
|
|
And 42 as server seed
|
|
And 2 slots
|
|
And 1024 as batch size
|
|
And 1024 as ubatch size
|
|
And 2048 KV cache size
|
|
And embeddings extraction
|
|
Then the server is starting
|
|
Then the server is healthy
|
|
|
|
Scenario: Embedding
|
|
When embeddings are computed for:
|
|
"""
|
|
What is the capital of Bulgaria ?
|
|
"""
|
|
Then embeddings are generated
|
|
|
|
Scenario: OAI Embeddings compatibility
|
|
Given a model bert-bge-small
|
|
When an OAI compatible embeddings computation request for:
|
|
"""
|
|
What is the capital of Spain ?
|
|
"""
|
|
Then embeddings are generated
|
|
|
|
Scenario: OAI Embeddings compatibility with multiple inputs
|
|
Given a model bert-bge-small
|
|
Given a prompt:
|
|
"""
|
|
In which country Paris is located ?
|
|
"""
|
|
And a prompt:
|
|
"""
|
|
Is Madrid the capital of Spain ?
|
|
"""
|
|
When an OAI compatible embeddings computation request for multiple inputs
|
|
Then embeddings are generated
|
|
|
|
Scenario: Multi users embeddings
|
|
Given a prompt:
|
|
"""
|
|
Write a very long story about AI.
|
|
"""
|
|
And a prompt:
|
|
"""
|
|
Write another very long music lyrics.
|
|
"""
|
|
And a prompt:
|
|
"""
|
|
Write a very long poem.
|
|
"""
|
|
And a prompt:
|
|
"""
|
|
Write a very long joke.
|
|
"""
|
|
Given concurrent embedding requests
|
|
Then the server is busy
|
|
Then the server is idle
|
|
Then all embeddings are generated
|
|
|
|
Scenario: Multi users OAI compatibility embeddings
|
|
Given a prompt:
|
|
"""
|
|
In which country Paris is located ?
|
|
"""
|
|
And a prompt:
|
|
"""
|
|
Is Madrid the capital of Spain ?
|
|
"""
|
|
And a prompt:
|
|
"""
|
|
What is the biggest US city ?
|
|
"""
|
|
And a prompt:
|
|
"""
|
|
What is the capital of Bulgaria ?
|
|
"""
|
|
And a model bert-bge-small
|
|
Given concurrent OAI embedding requests
|
|
Then the server is busy
|
|
Then the server is idle
|
|
Then all embeddings are generated
|
|
|
|
Scenario: All embeddings should be the same
|
|
Given 10 fixed prompts
|
|
And a model bert-bge-small
|
|
Given concurrent OAI embedding requests
|
|
Then all embeddings are the same
|