llama-cpp-haskell 0.3 → 0.3.0.1

raw patch · 3 files changed

+46/−26 lines, 3 filesPVP: minor bump suggested

API additions: PVP suggests at least a minor version bump

API changes (from Hackage documentation)

Files

• Llama.hs +37/−20
• Llama/Streaming.hs +6/−3
• llama-cpp-haskell.cabal +3/−3

@@ -88,9 +88,32 @@   } deriving (Show, Generic) instance FromJSON LlamaResponse +data LlamaError = LlamaError+  { code :: Word+  , message :: Text+  } deriving (Show, Generic)+instance FromJSON LlamaError++newtype LlamaResponseError = LlamaResponseError+  { error :: LlamaError+  } deriving (Show, Generic)+instance FromJSON LlamaResponseError+ type Token = Word32 type URL = String +llamaDecode :: (FromJSON a) => ByteString -> IO (Maybe a)+llamaDecode x =+  case decode x of+    Just v -> return v+    Nothing -> do+      case decode x of+        Just (LlamaResponseError err) -> do+          liftIO $ hPutStrLn stderr $ "llama-server returned an error: " ++ show err+        Nothing -> do+          liftIO $ hPutStrLn stderr $ "Failed to decode Llama response, got: " ++ show x+      return Nothing+ -- |Apply the LLM tempate to produce a raw LLM prompt from the role-content pairs applyTemplateSimple :: URL -> LlamaApplyTemplateRequest -> IO (Maybe Text) applyTemplateSimple = applyTemplateGeneral httpLBS@@ -109,11 +132,8 @@                     , requestHeaders = [("Content-Type", "application/json")]                     }   response <- fetch req-  case decode (responseBody response) of-    Just (LlamaApplyTemplateResponse text) -> return (Just text)-    Nothing -> do-      liftIO $ hPutStrLn stderr $ "Failed to decode Llama response, got: " ++ (show $ responseBody response)-      return Nothing+  decoded <- llamaDecode (responseBody response)+  return $ decoded >>= (\(LlamaApplyTemplateResponse text) -> Just text)  -- |Simple completion API sendToLlama :: URL -> Manager -> Text -> IO (Maybe Text)@@ -130,11 +150,8 @@                     , responseTimeout = responseTimeoutMicro 1800000000                     }   response <- httpLbs req manager-  case decode (responseBody response) of-    Just (LlamaResponse text) -> return (Just text)-    Nothing -> do-      liftIO $ hPutStrLn stderr $ "Failed to decode Llama response, got: " ++ (show $ responseBody response)-      return Nothing+  decoded <- llamaDecode (responseBody response)+  return $ decoded >>= (\(LlamaResponse text) -> Just text)  -- |Returns a token-by-token stream sendToLlamaStreaming :: (MonadThrow m, MonadResource m) => URL -> Manager -> Text -> IO (ConduitT () LlamaStreamingResponse m ())@@ -150,6 +167,12 @@                     , requestHeaders = [("Content-Type", "application/json")]                     }   pure $ httpSource req getResponseBody .| eventConduit+--  pure $ httpSource req getResponseBody .| debugConduit "stream" .| eventConduit+--+--debugConduit :: (MonadIO m, Show a) => String -> ConduitT a a m ()+--debugConduit prefix = awaitForever $ \x -> do+--    liftIO $ hPutStrLn stderr (prefix ++ ": " ++ show x)+--    yield x  tokenize :: URL -> LlamaTokenizeRequest -> IO (Maybe [Token]) tokenize url input = do@@ -159,11 +182,8 @@                     , requestHeaders = [("Content-Type", "application/json")]                     }   response <- httpLBS req-  case decode (responseBody response) of-    Just (LlamaTokenizeResponse result) -> return (Just result)-    Nothing -> do-      liftIO $ hPutStrLn stderr $ "Failed to decode Llama response, got: " ++ (show $ responseBody response)-      return Nothing+  decoded <- llamaDecode (responseBody response)+  return $ decoded >>= (\(LlamaTokenizeResponse result) -> Just result)  detokenize :: URL -> [Token] -> IO (Maybe Text) detokenize url input = do@@ -174,11 +194,8 @@                     , requestHeaders = [("Content-Type", "application/json")]                     }   response <- httpLBS req-  case decode (responseBody response) of-    Just (LlamaDetokenizeResponse text) -> return (Just text)-    Nothing -> do-      liftIO $ hPutStrLn stderr $ "Failed to decode Llama response, got: " ++ (show $ responseBody response)-      return Nothing+  decoded <- llamaDecode (responseBody response)+  return $ decoded >>= (\(LlamaDetokenizeResponse result) -> Just result)  -- |Extremely basic interface llama :: URL -> Text -> IO (Maybe Text)

@@ -58,7 +58,7 @@   { eventComment :: ByteString   } | RetryEvent   { eventRetry :: Int-  } | CloseEvent+  } | DispatchEvent   deriving (Show)  data LlamaStreamingResponse = LlamaStreamingResponse@@ -79,19 +79,22 @@  -- |text/event-stream parser event :: Parser ServerEvent-event = (sevent <|> comment <|> retry) <* eol+event = (sevent <|> comment <|> retry <|> dispatch) <* eol  sevent :: Parser ServerEvent sevent = ServerEvent   <$> optional (string "event" *> char ':' *> chars <* eol)   <*> optional (string "id"    *> char ':' *> chars <* eol)-  <*> many     (string "data"  *> char ':' *> chars <* eol)+  <*> many1    (string "data"  *> char ':' *> chars <* eol) -- technically SSE can have no data, but it is not the case for llama-server  comment :: Parser ServerEvent comment = CommentEvent <$> (char ':' *> chars <* eol)  retry :: Parser ServerEvent retry = RetryEvent <$> (string "retry:" *> decimal <* eol)++dispatch :: Parser ServerEvent+dispatch = pure DispatchEvent  chars :: Parser ByteString chars = AC8.takeTill (== '\n')

@@ -1,6 +1,6 @@ cabal-version:      2.2 name:               llama-cpp-haskell-version:            0.3+version:            0.3.0.1 synopsis:           Haskell bindings for the llama.cpp llama-server and a simple CLI description:        This is the interface that allows one to interface with llama-server RPC API using Haskell concepts. It also includes a `llamacall` binary to do it from your favorite command line shell and use it in scripting. license:            AGPL-3.0-only@@ -10,7 +10,7 @@ -- copyright: category:           Text, LLM, Llama, Machine Learning, AI, Network, CLI build-type:         Simple-tested-with:        GHC == 9.12.2+tested-with:        GHC == 9.12.3 -- extra-source-files:  Source-repository head@@ -20,7 +20,7 @@ Source-repository this   type:              git   location:          https://github.com/l29ah/llama-cpp-haskell.git-  tag:               0.3+  tag:               0.3.0.1  common stuff     ghc-options: -Wall