still not really working right

This commit is contained in:
LostRuins Concedo
2025-11-09 01:57:48 +08:00
48 changed files with 3193 additions and 2705 deletions
+6 -6
View File
@@ -77,10 +77,10 @@ def test_different_draft_min_draft_max():
def test_slot_ctx_not_exceeded():
global server
server.n_ctx = 64
server.n_ctx = 256
server.start()
res = server.make_request("POST", "/completion", data={
"prompt": "Hello " * 56,
"prompt": "Hello " * 248,
"temperature": 0.0,
"top_k": 1,
"speculative.p_min": 0.0,
@@ -91,19 +91,19 @@ def test_slot_ctx_not_exceeded():
def test_with_ctx_shift():
global server
server.n_ctx = 64
server.n_ctx = 256
server.enable_ctx_shift = True
server.start()
res = server.make_request("POST", "/completion", data={
"prompt": "Hello " * 56,
"prompt": "Hello " * 248,
"temperature": 0.0,
"top_k": 1,
"n_predict": 64,
"n_predict": 256,
"speculative.p_min": 0.0,
})
assert res.status_code == 200
assert len(res.body["content"]) > 0
assert res.body["tokens_predicted"] == 64
assert res.body["tokens_predicted"] == 256
assert res.body["truncated"] == True