mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-08-12 22:31:11 +04:00
dd1ea52433
* Enable backend sampling with token speculation * Clamp the mask sum before converting it into the sampled index * Add a numeric context parameter declaring the maximum outputs one sequence * More fixes * Don't reuse memory for output views. * Match dist between CPU and GPU * Fix CPU and backend sampling mismatches * Simpify some of the changes * Fix tests on Vulkan * More test fixes * Rebase changes * Rebase and address review comments * Address review comments * Address review comments * Update src/llama-sampler.cpp Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
136 lines
3.7 KiB
Python
136 lines
3.7 KiB
Python
import pytest
|
|
from utils import *
|
|
|
|
# We use a F16 MOE gguf as main model, and q4_0 as draft model
|
|
|
|
server = ServerPreset.stories15m_moe()
|
|
|
|
MODEL_DRAFT_FILE_URL = "https://huggingface.co/ggml-org/tiny-llamas/resolve/main/stories15M-q4_0.gguf"
|
|
|
|
def create_server():
|
|
global server
|
|
server = ServerPreset.stories15m_moe()
|
|
# set default values
|
|
server.model_draft = download_file(MODEL_DRAFT_FILE_URL)
|
|
server.spec_type = "draft-simple"
|
|
server.spec_draft_n_min = 4
|
|
server.spec_draft_n_max = 8
|
|
server.fa = "off"
|
|
|
|
|
|
@pytest.fixture(autouse=True)
|
|
def fixture_create_server():
|
|
return create_server()
|
|
|
|
|
|
def test_with_and_without_draft():
|
|
global server
|
|
request = {
|
|
"prompt": "I believe the meaning of life is",
|
|
"temperature": 0.8,
|
|
"top_k": 40,
|
|
"seed": 4242,
|
|
"n_predict": 16,
|
|
"return_tokens": True,
|
|
}
|
|
|
|
server.model_draft = None # disable draft model
|
|
server.spec_type = None
|
|
server.backend_sampling = True
|
|
server.start()
|
|
res = server.make_request("POST", "/completion", data=request)
|
|
assert res.status_code == 200
|
|
tokens_no_draft = res.body["tokens"]
|
|
server.stop()
|
|
|
|
# create new server with draft model
|
|
create_server()
|
|
server.backend_sampling = True
|
|
server.start()
|
|
res = server.make_request("POST", "/completion", data=request)
|
|
assert res.status_code == 200
|
|
assert res.body["timings"]["draft_n"] > 0
|
|
tokens_draft = res.body["tokens"]
|
|
|
|
assert tokens_no_draft == tokens_draft
|
|
|
|
|
|
def test_different_draft_min_draft_max():
|
|
global server
|
|
test_values = [
|
|
(1, 2),
|
|
(1, 4),
|
|
(4, 8),
|
|
(4, 12),
|
|
(8, 16),
|
|
]
|
|
last_content = None
|
|
for draft_min, draft_max in test_values:
|
|
server.stop()
|
|
server.spec_draft_n_min = draft_min
|
|
server.spec_draft_n_max = draft_max
|
|
server.start()
|
|
res = server.make_request("POST", "/completion", data={
|
|
"prompt": "I believe the meaning of life is",
|
|
"temperature": 0.0,
|
|
"top_k": 1,
|
|
"n_predict": 16,
|
|
})
|
|
assert res.status_code == 200
|
|
if last_content is not None:
|
|
assert last_content == res.body["content"]
|
|
last_content = res.body["content"]
|
|
|
|
|
|
def test_slot_ctx_not_exceeded():
|
|
global server
|
|
server.n_ctx = 256
|
|
server.start()
|
|
res = server.make_request("POST", "/completion", data={
|
|
"prompt": "Hello " * 248,
|
|
"temperature": 0.0,
|
|
"top_k": 1,
|
|
"speculative.p_min": 0.0,
|
|
})
|
|
assert res.status_code == 200
|
|
assert len(res.body["content"]) > 0
|
|
|
|
|
|
def test_with_ctx_shift():
|
|
global server
|
|
server.n_ctx = 256
|
|
server.enable_ctx_shift = True
|
|
server.start()
|
|
res = server.make_request("POST", "/completion", data={
|
|
"prompt": "Hello " * 248,
|
|
"temperature": 0.0,
|
|
"top_k": 1,
|
|
"n_predict": 256,
|
|
"speculative.p_min": 0.0,
|
|
})
|
|
assert res.status_code == 200
|
|
assert len(res.body["content"]) > 0
|
|
assert res.body["tokens_predicted"] == 256
|
|
assert res.body["truncated"] == True
|
|
|
|
|
|
@pytest.mark.parametrize("n_slots,n_requests", [
|
|
(1, 2),
|
|
(2, 2),
|
|
])
|
|
def test_multi_requests_parallel(n_slots: int, n_requests: int):
|
|
global server
|
|
server.n_slots = n_slots
|
|
server.start()
|
|
tasks = []
|
|
for _ in range(n_requests):
|
|
tasks.append((server.make_request, ("POST", "/completion", {
|
|
"prompt": "I believe the meaning of life is",
|
|
"temperature": 0.0,
|
|
"top_k": 1,
|
|
})))
|
|
results = parallel_function_calls(tasks)
|
|
for res in results:
|
|
assert res.status_code == 200
|
|
assert match_regex("(wise|kind|owl|answer)+", res.body["content"])
|