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

default horde name and better handling for horde (+3 squashed commit)

Browse source 
Squashed commit:

[fadfa60] better idle handling for horde worker

[a3971e6] updated lite

[2ca2b79] seems to not generate rubbish
This commit is contained in:
Concedo 2023-07-24 17:37:26 +08:00
parent c7136f03d9
commit 825e34baa3
3 changed files with 10 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

2
CMakeLists.txt
View file

@ -99,7 +99,7 @@ if (LLAMA_CUBLAS)
if (LLAMA_CUDA_DMMV_F16)
set(CMAKE_CUDA_ARCHITECTURES "60;61") # needed for f16 CUDA intrinsics
else()
set(CMAKE_CUDA_ARCHITECTURES "37;86") # lowest CUDA 12 standard + lowest for integer intrinsics
set(CMAKE_CUDA_ARCHITECTURES "37;61") # lowest CUDA 12 standard + lowest for integer intrinsics
endif()
endif()
message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")

2
klite.embd
View file

File diff suppressed because one or more lines are too long

11
koboldcpp.py
View file

@ -978,7 +978,7 @@ def show_new_gui():
else:
item.grid_forget()
labels[idx].grid_forget()
if usehorde_var.get()==1 and horde_name_var.get()=="koboldcpp" and model_var.get()!="":
if usehorde_var.get()==1 and (horde_name_var.get()=="koboldcpp" or horde_name_var.get()=="") and model_var.get()!="":
basefile = os.path.basename(model_var.get())
horde_name_var.set(os.path.splitext(basefile)[0])
@ -1388,6 +1388,7 @@ def run_horde_worker(args, api_key, worker_name):
current_id = None
current_payload = None
current_generation = None
sleepy_counter = 0 #if this exceeds a value, worker becomes sleepy (slower)
print("===\nEmbedded Horde Worker '"+worker_name+"' Starting...\n(To use your own KAI Bridge/Scribe worker instead, don't set your API key)")
BRIDGE_AGENT = f"KoboldCppEmbedWorker:1:https://github.com/LostRuins/koboldcpp"
cluster = "https://horde.koboldai.net"
@ -1424,9 +1425,13 @@ def run_horde_worker(args, api_key, worker_name):
time.sleep(5)
continue
if not pop["id"]:
#print(f"Server {cluster} has no valid generations to do for us.")
time.sleep(3)
slp = (2 if sleepy_counter<10 else (3 if sleepy_counter<20 else 4))
#print(f"Server {cluster} has no valid generations for us. Sleep for {slp}s")
time.sleep(slp)
sleepy_counter += 1
continue
sleepy_counter = 0
current_id = pop['id']
current_payload = pop['payload']
print(f"\nJob received from {cluster} for {current_payload.get('max_length',80)} tokens and {current_payload.get('max_context_length',1024)} max context. Starting generation...")