github/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2025-06-26 15:31:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix regenerate button to be deterministic and bump the llama version to latest we have for gguf.

Browse Source
This commit is contained in:
Adam Treat 2023-10-03 12:42:31 -04:00
parent 40c78d2f78
commit 12f943e966
4 changed files with 5 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
gpt4all-backend/llama.cpp-mainline

@ -1 +1 @@
Subproject commit 37a0be313d21f8b61184a3adcaac123353128238 Subproject commit 70a6537c4aae9951ba7fff740135ca7dbe14d0f1

9
gpt4all-backend/llamamodel.cpp
View File

@ -249,15 +249,6 @@ LLModel::Token LLamaModel::sampleToken(PromptContext &promptCtx) const
bool LLamaModel::evalTokens(PromptContext &ctx, const std::vector<int32_t> &tokens) const bool LLamaModel::evalTokens(PromptContext &ctx, const std::vector<int32_t> &tokens) const
{ {
// When we recalculate context we could have erased the original BOS token... we need to replace it
const bool useBOS = ctx.n_past == 0 && (ctx.tokens.empty() || ctx.tokens.front() != llama_token_bos(d_ptr->ctx));
if (useBOS) {
std::vector<int32_t> myTokens;
myTokens.push_back(llama_token_bos(d_ptr->ctx));
myTokens.insert(myTokens.end(), tokens.begin(), tokens.end());
ctx.n_past += 1;
return llama_eval(d_ptr->ctx, myTokens.data(), myTokens.size(), ctx.n_past, d_ptr->n_threads) == 0;
} else
return llama_eval(d_ptr->ctx, tokens.data(), tokens.size(), ctx.n_past, d_ptr->n_threads) == 0; return llama_eval(d_ptr->ctx, tokens.data(), tokens.size(), ctx.n_past, d_ptr->n_threads) == 0;
} }

5
gpt4all-backend/llmodel_shared.cpp
View File

@ -92,10 +92,10 @@ void LLModel::prompt(const std::string &prompt,
if (int32_t(promptCtx.tokens.size()) == promptCtx.n_ctx) if (int32_t(promptCtx.tokens.size()) == promptCtx.n_ctx)
promptCtx.tokens.erase(promptCtx.tokens.begin()); promptCtx.tokens.erase(promptCtx.tokens.begin());
promptCtx.tokens.push_back(batch.at(t)); promptCtx.tokens.push_back(batch.at(t));
promptCtx.n_past += 1;
if (!promptCallback(batch.at(t))) if (!promptCallback(batch.at(t)))
return; return;
} }
promptCtx.n_past += batch.size();
i = batch_end; i = batch_end;
} }
@ -126,8 +126,6 @@ void LLModel::prompt(const std::string &prompt,
return; return;
} }
promptCtx.n_past += 1;
// display text // display text
for (const auto token : endTokens()) { for (const auto token : endTokens()) {
if (id == token) return; if (id == token) return;
@ -162,6 +160,7 @@ void LLModel::prompt(const std::string &prompt,
if (int32_t(promptCtx.tokens.size()) == promptCtx.n_ctx) if (int32_t(promptCtx.tokens.size()) == promptCtx.n_ctx)
promptCtx.tokens.erase(promptCtx.tokens.begin()); promptCtx.tokens.erase(promptCtx.tokens.begin());
promptCtx.tokens.push_back(t); promptCtx.tokens.push_back(t);
promptCtx.n_past += 1;
//TODO: Conversion to std::string can be avoided here... //TODO: Conversion to std::string can be avoided here...
if (!responseCallback(t, std::string(tokenToString(t)))) if (!responseCallback(t, std::string(tokenToString(t))))
return; return;

2
gpt4all-chat/chatllm.cpp
View File

@ -371,7 +371,7 @@ void ChatLLM::regenerateResponse()
else else
m_ctx.n_past -= m_promptResponseTokens; m_ctx.n_past -= m_promptResponseTokens;
m_ctx.n_past = std::max(0, m_ctx.n_past); m_ctx.n_past = std::max(0, m_ctx.n_past);
m_ctx.tokens.erase(m_ctx.tokens.end() -= m_promptResponseTokens, m_ctx.tokens.end()); m_ctx.tokens.erase(m_ctx.tokens.end() - m_promptResponseTokens, m_ctx.tokens.end());
m_promptResponseTokens = 0; m_promptResponseTokens = 0;
m_promptTokens = 0; m_promptTokens = 0;
m_response = std::string(); m_response = std::string();