Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -3,6 +3,7 @@ import os, json, asyncio, torch
|
|
3 |
from fastapi import FastAPI, WebSocket, WebSocketDisconnect
|
4 |
from huggingface_hub import login
|
5 |
from transformers import AutoModelForCausalLM, AutoTokenizer, LogitsProcessor
|
|
|
6 |
from snac import SNAC
|
7 |
|
8 |
# ββ 0.Β HFβAuth & Device ββββββββββββββββββββββββββββββββββββββββββββββ
|
@@ -114,7 +115,10 @@ async def tts(ws: WebSocket):
|
|
114 |
use_cache=True,
|
115 |
return_dict_in_generate=True,
|
116 |
)
|
117 |
-
|
|
|
|
|
|
|
118 |
newtok = out.sequences[0,-out.num_generated_tokens:].tolist()
|
119 |
|
120 |
for t in newtok:
|
|
|
3 |
from fastapi import FastAPI, WebSocket, WebSocketDisconnect
|
4 |
from huggingface_hub import login
|
5 |
from transformers import AutoModelForCausalLM, AutoTokenizer, LogitsProcessor
|
6 |
+
from transformers.generation.utils import Cache
|
7 |
from snac import SNAC
|
8 |
|
9 |
# ββ 0.Β HFβAuth & Device ββββββββββββββββββββββββββββββββββββββββββββββ
|
|
|
115 |
use_cache=True,
|
116 |
return_dict_in_generate=True,
|
117 |
)
|
118 |
+
pkv = out.past_key_values
|
119 |
+
if isinstance(pkv, Cache):
|
120 |
+
pkv = pkv.to_legacy()
|
121 |
+
past_kvs = pkv
|
122 |
newtok = out.sequences[0,-out.num_generated_tokens:].tolist()
|
123 |
|
124 |
for t in newtok:
|