DSPy: The framework for programming—not prompting—language models
Python
Allow passing of kwargs for HF TGI request
Guy Aglionby516d45da0131
diff --git a/dsp/modules/hf_client.py b/dsp/modules/hf_client.pyindex 1ff3d1bb2..bd2061e7a 100644--- a/dsp/modules/hf_client.py+++ b/dsp/modules/hf_client.py@@ -9,11 +9,12 @@ from dsp.modules.cache_utils import CacheMemory, NotebookCacheMemory, cache_turnclass HFClientTGI(HFModel):- def __init__(self, model, port, url="http://future-hgx-1", **kwargs):+ def __init__(self, model, port, url="http://future-hgx-1", http_request_kwargs=None, **kwargs):super().__init__(model=model, is_client=True)self.url = urlself.ports = port if isinstance(port, list) else [port]+ self.http_request_kwargs = http_request_kwargs or {}self.headers = {"Content-Type": "application/json"}@@ -53,7 +54,14 @@ class HFClientTGI(HFModel):# response = requests.post(self.url + "/generate", json=payload, headers=self.headers)- response = send_hftgi_request_v01_wrapped(f"{self.url}:{random.Random().choice(self.ports)}" + "/generate", url=self.url, ports=tuple(self.ports), json=payload, headers=self.headers)+ response = send_hftgi_request_v01_wrapped(+ f"{self.url}:{random.Random().choice(self.ports)}" + "/generate",+ url=self.url,+ ports=tuple(self.ports),+ json=payload,+ headers=self.headers,+ **self.http_request_kwargs,+ )try:json_response = response.json()