get rid of some more hf bloat

Author: AlexCheema

exo/download/test_new_shard_download.py

@@ -1,4 +1,4 @@
-from exo.download.hf.new_shard_download import download_shard, NewShardDownloader
+from exo.download.new_shard_download import download_shard, NewShardDownloader
 from exo.inference.shard import Shard
 from exo.models import get_model_id
 from pathlib import Path

exo/inference/mlx/test_non_blocking.py

@@ -2,15 +2,15 @@
 import time
 import numpy as np
 from exo.inference.mlx.sharded_inference_engine import MLXDynamicShardInferenceEngine
-from exo.download.hf.hf_shard_download import HFShardDownloader
+from exo.download.new_shard_download import NewShardDownloader
 from exo.inference.shard import Shard
 from exo.models import build_base_shard
 from collections import deque
 from statistics import mean, median
 
 async def test_non_blocking():
     # Setup
-    shard_downloader = HFShardDownloader()
+    shard_downloader = NewShardDownloader()
     engine = MLXDynamicShardInferenceEngine(shard_downloader)
     _shard = build_base_shard("llama-3.1-8b", "MLXDynamicShardInferenceEngine")
     shard = Shard(_shard.model_id, _shard.start_layer, _shard.n_layers - 1, _shard.n_layers)

exo/inference/test_inference_engine.py

@@ -1,6 +1,6 @@
 from exo.inference.mlx.sharded_inference_engine import MLXDynamicShardInferenceEngine
-from exo.download.hf.hf_shard_download import HFShardDownloader
 from exo.inference.inference_engine import InferenceEngine
+from exo.download.new_shard_download import NewShardDownloader
 from exo.inference.shard import Shard
 from exo.helpers import DEBUG
 import os
@@ -44,13 +44,13 @@ async def test_inference_engine(inference_engine_1: InferenceEngine, inference_e
   assert np.array_equal(next_resp_full, resp4)
 
 
-asyncio.run(test_inference_engine(MLXDynamicShardInferenceEngine(HFShardDownloader()), MLXDynamicShardInferenceEngine(HFShardDownloader()), "llama-3.2-1b", 16))
+asyncio.run(test_inference_engine(MLXDynamicShardInferenceEngine(NewShardDownloader()), MLXDynamicShardInferenceEngine(NewShardDownloader()), "llama-3.2-1b", 16))
 
 if os.getenv("RUN_TINYGRAD", default="0") == "1":
   import tinygrad
   import os
   from exo.inference.tinygrad.inference import TinygradDynamicShardInferenceEngine
   tinygrad.helpers.DEBUG.value = int(os.getenv("TINYGRAD_DEBUG", default="0"))
   asyncio.run(
-    test_inference_engine(TinygradDynamicShardInferenceEngine(HFShardDownloader()), TinygradDynamicShardInferenceEngine(HFShardDownloader()), "llama-3-8b", 32)
+    test_inference_engine(TinygradDynamicShardInferenceEngine(NewShardDownloader()), TinygradDynamicShardInferenceEngine(NewShardDownloader()), "llama-3-8b", 32)
   )