Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 01:05:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

fixed save state a bit better

Browse Source
This commit is contained in:
Concedo
2025-12-21 22:24:13 +08:00
parent b51e3592ba
commit 4b899b19dc
1 changed files with 13 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gpttype_adapter.cpp
+13
View File
@@ -4111,6 +4111,10 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
n_past -=1;
embd_inp.insert(embd_inp.begin(), 1, tail);
}
else if(maxedpos==n_past)
{
n_past += 1;
}
}
}
}
@@ -5036,6 +5040,15 @@ size_t gpttype_save_state_kv(int slot)
totalbytes += res;
savestates[slot].current_savestate_size = newsize;
savestates[slot].savestate_context_tokens = current_context_tokens;
int maxedpos = llama_memory_seq_pos_max(llama_get_memory(llama_ctx_v4),0);
if(maxedpos > 0 && savestates[slot].savestate_context_tokens.size() > maxedpos && savestates[slot].savestate_context_tokens.size()-maxedpos<=2)
{
//dirty hack for the memory actually being off by 1 or 2, correct the state
while(savestates[slot].savestate_context_tokens.size() > maxedpos)
{
savestates[slot].savestate_context_tokens.pop_back();
}
}
touch_slot(slot);
printf("\nKV Save State %d: Created SaveState of %zu tokens, costing %zu MB.\n",slot,current_context_tokens.size(),savestates[slot].current_savestate_size/(1024*1024));
}