2024-02-24 12:28:55 +01:00
|
|
|
@llama.cpp
|
2024-03-02 22:00:14 +01:00
|
|
|
@parallel
|
2024-02-24 12:28:55 +01:00
|
|
|
Feature: Parallel
|
|
|
|
|
|
|
|
Background: Server startup
|
|
|
|
Given a server listening on localhost:8080
|
2024-03-23 18:07:00 +01:00
|
|
|
And a model file tinyllamas/split/stories15M-00001-of-00003.gguf from HF repo ggml-org/models
|
|
|
|
And a model file test-model-00001-of-00003.gguf
|
2024-02-24 12:28:55 +01:00
|
|
|
And 42 as server seed
|
2024-03-09 10:30:04 +01:00
|
|
|
And 128 as batch size
|
|
|
|
And 256 KV cache size
|
2024-02-24 12:28:55 +01:00
|
|
|
And 2 slots
|
|
|
|
And continuous batching
|
|
|
|
Then the server is starting
|
|
|
|
Then the server is healthy
|
|
|
|
|
|
|
|
Scenario Outline: Multi users completion
|
|
|
|
Given a prompt:
|
|
|
|
"""
|
|
|
|
Write a very long story about AI.
|
|
|
|
"""
|
|
|
|
And a prompt:
|
|
|
|
"""
|
|
|
|
Write another very long music lyrics.
|
|
|
|
"""
|
|
|
|
And <n_predict> max tokens to predict
|
|
|
|
Given concurrent completion requests
|
|
|
|
Then the server is busy
|
|
|
|
Then the server is idle
|
|
|
|
And all slots are idle
|
|
|
|
Then all prompts are predicted with <n_predict> tokens
|
|
|
|
Examples:
|
|
|
|
| n_predict |
|
|
|
|
| 128 |
|
|
|
|
|
|
|
|
Scenario Outline: Multi users OAI completions compatibility
|
|
|
|
Given a system prompt You are a writer.
|
|
|
|
And a model tinyllama-2
|
|
|
|
Given a prompt:
|
|
|
|
"""
|
|
|
|
Write a very long book.
|
|
|
|
"""
|
|
|
|
And a prompt:
|
|
|
|
"""
|
|
|
|
Write another a poem.
|
|
|
|
"""
|
|
|
|
And <n_predict> max tokens to predict
|
|
|
|
And streaming is <streaming>
|
|
|
|
Given concurrent OAI completions requests
|
|
|
|
Then the server is busy
|
|
|
|
Then the server is idle
|
|
|
|
Then all prompts are predicted with <n_predict> tokens
|
|
|
|
Examples:
|
|
|
|
| streaming | n_predict |
|
2024-09-02 22:08:38 +02:00
|
|
|
| disabled | 128 |
|
|
|
|
| enabled | 64 |
|
2024-02-24 12:28:55 +01:00
|
|
|
|
2024-02-28 09:39:15 +01:00
|
|
|
Scenario Outline: Multi users OAI completions compatibility no v1
|
|
|
|
Given a system prompt You are a writer.
|
|
|
|
And a model tinyllama-2
|
|
|
|
Given a prompt:
|
|
|
|
"""
|
|
|
|
Write a very long book.
|
|
|
|
"""
|
|
|
|
And a prompt:
|
|
|
|
"""
|
|
|
|
Write another a poem.
|
|
|
|
"""
|
|
|
|
And <n_predict> max tokens to predict
|
|
|
|
And streaming is <streaming>
|
|
|
|
Given concurrent OAI completions requests no v1
|
|
|
|
Then the server is busy
|
|
|
|
Then the server is idle
|
|
|
|
Then all prompts are predicted with <n_predict> tokens
|
|
|
|
Examples:
|
|
|
|
| streaming | n_predict |
|
|
|
|
| disabled | 128 |
|
|
|
|
| enabled | 64 |
|
|
|
|
|
2024-09-06 23:21:29 +02:00
|
|
|
Scenario Outline: Multi users with number of prompts exceeding number of slots
|
|
|
|
Given a system prompt You are a writer.
|
|
|
|
And a model tinyllama-2
|
|
|
|
Given a prompt:
|
|
|
|
"""
|
|
|
|
Write a very long book.
|
|
|
|
"""
|
|
|
|
And a prompt:
|
|
|
|
"""
|
|
|
|
Write another a poem.
|
|
|
|
"""
|
|
|
|
And a prompt:
|
|
|
|
"""
|
|
|
|
What is LLM?
|
|
|
|
"""
|
|
|
|
And a prompt:
|
|
|
|
"""
|
|
|
|
The sky is blue and I love it.
|
|
|
|
"""
|
|
|
|
And <n_predict> max tokens to predict
|
|
|
|
And streaming is <streaming>
|
|
|
|
Given concurrent OAI completions requests
|
|
|
|
Then the server is busy
|
|
|
|
Then the server is idle
|
|
|
|
Then all prompts are predicted with <n_predict> tokens
|
|
|
|
Examples:
|
|
|
|
| streaming | n_predict |
|
|
|
|
| disabled | 128 |
|
|
|
|
| enabled | 64 |
|
2024-03-09 10:30:04 +01:00
|
|
|
|
2024-02-24 12:28:55 +01:00
|
|
|
Scenario: Multi users with total number of tokens to predict exceeds the KV Cache size #3969
|
|
|
|
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.
|
|
|
|
"""
|
|
|
|
And 128 max tokens to predict
|
|
|
|
Given concurrent completion requests
|
|
|
|
Then the server is busy
|
|
|
|
Then the server is idle
|
|
|
|
Then all prompts are predicted
|