vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

include content-length

Browse source
This commit is contained in:
Concedo 2023-10-28 14:24:37 +08:00
parent 64f3bc5168
commit 09c74ea046
1 changed files with 19 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

39
koboldcpp.py
View file

@ -675,23 +675,22 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
content_length = int(self.headers['content-length']) content_length = int(self.headers['content-length'])
body = self.rfile.read(content_length) body = self.rfile.read(content_length)
self.path = self.path.rstrip('/') self.path = self.path.rstrip('/')
response_body = None
response_code = 200
if self.path.endswith(('/api/extra/tokencount')): if self.path.endswith(('/api/extra/tokencount')):
try: try:
genparams = json.loads(body) genparams = json.loads(body)
countprompt = genparams.get('prompt', "") countprompt = genparams.get('prompt', "")
count = handle.token_count(countprompt.encode("UTF-8")) count = handle.token_count(countprompt.encode("UTF-8"))
self.send_response(200) response_body = (json.dumps({"value": count}).encode())
self.end_headers(content_type='application/json')
self.wfile.write(json.dumps({"value": count}).encode())
except Exception as e: except Exception as e:
utfprint("Count Tokens - Body Error: " + str(e)) utfprint("Count Tokens - Body Error: " + str(e))
self.send_response(400) response_code = 400
self.end_headers(content_type='application/json') response_body = (json.dumps({"value": -1}).encode())
self.wfile.write(json.dumps({"value": -1}).encode())
return
if self.path.endswith('/api/extra/abort'): elif self.path.endswith('/api/extra/abort'):
multiuserkey = "" multiuserkey = ""
try: try:
tempbody = json.loads(body) tempbody = json.loads(body)
@ -704,17 +703,12 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
if (multiuserkey=="" and requestsinqueue==0) or (multiuserkey!="" and multiuserkey==currentusergenkey): if (multiuserkey=="" and requestsinqueue==0) or (multiuserkey!="" and multiuserkey==currentusergenkey):
ag = handle.abort_generate() ag = handle.abort_generate()
time.sleep(0.1) #short delay before replying time.sleep(0.1) #short delay before replying
self.send_response(200) response_body = (json.dumps({"success": ("true" if ag else "false")}).encode())
self.end_headers(content_type='application/json')
self.wfile.write(json.dumps({"success": ("true" if ag else "false")}).encode())
print("\nGeneration Aborted") print("\nGeneration Aborted")
else: else:
self.send_response(200) response_body = (json.dumps({"success": "false"}).encode())
self.end_headers(content_type='application/json')
self.wfile.write(json.dumps({"success": "false"}).encode())
return
if self.path.endswith('/api/extra/generate/check'): elif self.path.endswith('/api/extra/generate/check'):
pendtxtStr = "" pendtxtStr = ""
multiuserkey = "" multiuserkey = ""
try: try:
@ -723,15 +717,18 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
multiuserkey = tempbody.get('genkey', "") multiuserkey = tempbody.get('genkey', "")
except Exception as e: except Exception as e:
multiuserkey = "" multiuserkey = ""
pass
if totalgens>0: if totalgens>0:
if (multiuserkey=="" and requestsinqueue==0) or (multiuserkey!="" and multiuserkey==currentusergenkey): if (multiuserkey=="" and requestsinqueue==0) or (multiuserkey!="" and multiuserkey==currentusergenkey):
pendtxt = handle.get_pending_output() pendtxt = handle.get_pending_output()
pendtxtStr = ctypes.string_at(pendtxt).decode("UTF-8","ignore") pendtxtStr = ctypes.string_at(pendtxt).decode("UTF-8","ignore")
self.send_response(200) response_body = (json.dumps({"results": [{"text": pendtxtStr}]}).encode())
if response_body is not None:
self.send_response(response_code)
self.send_header('content-length', str(len(response_body)))
self.end_headers(content_type='application/json') self.end_headers(content_type='application/json')
self.wfile.write(json.dumps({"results": [{"text": pendtxtStr}]}).encode()) self.wfile.write(response_body)
return return
reqblocking = False reqblocking = False
@ -794,8 +791,10 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
# Headers are already sent when streaming # Headers are already sent when streaming
if not sse_stream_flag: if not sse_stream_flag:
self.send_response(200) self.send_response(200)
genresp = (json.dumps(gen).encode())
self.send_header('content-length', str(len(genresp)))
self.end_headers(content_type='application/json') self.end_headers(content_type='application/json')
self.wfile.write(json.dumps(gen).encode()) self.wfile.write(genresp)
except: except:
print("Generate: The response could not be sent, maybe connection was terminated?") print("Generate: The response could not be sent, maybe connection was terminated?")
return return