Merge pull request #264 from exo-explore/unused_deps

remove hf-transfer, huggingface-hub, Jinja2 unused dependencies

exo/inference/tinygrad/inference.py

@@ -4,9 +4,8 @@ import os
 from exo.inference.tinygrad.models.llama import Transformer, convert_from_huggingface, fix_bf16
 from exo.inference.shard import Shard
 from exo.inference.tokenizers import resolve_tokenizer
-from tinygrad.nn.state import safe_load, torch_load, load_state_dict
-from tinygrad import Tensor, dtypes, nn, Context
-from transformers import AutoTokenizer
+from tinygrad.nn.state import load_state_dict
+from tinygrad import Tensor, nn, Context
 from exo.inference.inference_engine import InferenceEngine
 from typing import Optional, Tuple
 import numpy as np
@@ -14,8 +13,6 @@ from exo.inference.tinygrad.tinygrad_helpers import concat_weights, load
 from exo.download.shard_download import ShardDownloader
 from concurrent.futures import ThreadPoolExecutor
 import asyncio
-import threading
-from functools import partial
 
 Tensor.no_grad = True
 # default settings

setup.py

@@ -7,12 +7,8 @@ install_requires = [
   "aiohttp==3.10.2",
   "aiohttp_cors==0.7.0",
   "aiofiles==24.1.0",
-  "blobfile==2.1.1",
   "grpcio==1.64.1",
   "grpcio-tools==1.64.1",
-  "hf-transfer==0.1.8",
-  "huggingface-hub==0.24.5",
-  "Jinja2==3.1.4",
   "netifaces==0.11.0",
   "numpy==2.0.0",
   "pillow==10.4.0",
@@ -25,8 +21,6 @@ install_requires = [
   "safetensors==0.4.3",
   "tailscale==0.6.1",
   "tenacity==9.0.0",
-  "tiktoken==0.7.0",
-  "tokenizers==0.19.1",
   "tqdm==4.66.4",
   "transformers==4.43.3",
   "uuid==1.30",