test: align QDC tests to the precision API

Author: RemiliaForever

tests/conftest.py

@@ -117,7 +117,7 @@ def geniex_session():
 @pytest.fixture(scope='session')
 def llama_cpp_llm_paths(geniex_session):
     try:
-        return _mm.ensure_cached(LLAMA_CPP_LLM_MODEL, quant=LLAMA_CPP_LLM_QUANT, hub='hf')
+        return _mm.ensure_cached(LLAMA_CPP_LLM_MODEL, precision=LLAMA_CPP_LLM_QUANT, hub='hf')
     except geniex.GenieXError as e:
         pytest.skip(f'could not pull {LLAMA_CPP_LLM_MODEL}: {e}')
 

tests/plugins/llama_cpp/test_llama_cpp_llm.py

@@ -34,7 +34,7 @@
 def test_generate_blocking(llama_cpp_llm_paths, device_map):
     with geniex.AutoModelForCausalLM.from_pretrained(
         LLAMA_CPP_LLM_MODEL,
-        quant=LLAMA_CPP_LLM_QUANT,
+        precision=LLAMA_CPP_LLM_QUANT,
         device_map=device_map,
     ) as llm:
         assert isinstance(llm, geniex.GenieXLLM)
@@ -48,7 +48,7 @@ def test_generate_blocking(llama_cpp_llm_paths, device_map):
 def test_generate_stream(llama_cpp_llm_paths, device_map):
     with geniex.AutoModelForCausalLM.from_pretrained(
         LLAMA_CPP_LLM_MODEL,
-        quant=LLAMA_CPP_LLM_QUANT,
+        precision=LLAMA_CPP_LLM_QUANT,
         device_map=device_map,
     ) as llm:
         streamer = llm.generate('Say hi.', max_new_tokens=8, temperature=0.0, seed=42, stream=True)
@@ -70,7 +70,7 @@ def test_quality_keywords(llama_cpp_llm_paths, device_map, prompt, expected):
     # in completion mode and the keyword only appears by sampler luck.
     with geniex.AutoModelForCausalLM.from_pretrained(
         LLAMA_CPP_LLM_MODEL,
-        quant=LLAMA_CPP_LLM_QUANT,
+        precision=LLAMA_CPP_LLM_QUANT,
         device_map=device_map,
     ) as llm:
         formatted = llm.tokenizer.apply_chat_template(