Whisper: added support for custom models

examples/README.md

@@ -0,0 +1,5 @@
+# An example collection of gpnvim configurations
+
+## Content:
+
+- [Groq based config](./groq_gp_config.lua)

examples/groq_gp_config.lua

@@ -0,0 +1,90 @@
+local OPENAI_KEY = "<your_key>"
+local GROQ_KEY = "<your_key>"
+local OPENAI_HOST = "https://api.openai.com/v1/chat/completions"
+local GROQ_HOST = "https://api.groq.com/openai/v1/chat/completions"
+local GROQ_AUDIO = "https://api.groq.com/openai/v1/audio/transcriptions"
+
+local GROQ_WHISPER_MODEL = "distil-whisper-large-v3-en";
+
+-- Gp (GPT prompt) lua plugin for Neovim
+-- https://github.com/Robitx/gp.nvim/
+
+--------------------------------------------------------------------------------
+-- Default config
+--------------------------------------------------------------------------------
+---@class GpConfig
+-- README_REFERENCE_MARKER_START
+local config = {
+  providers = {
+    openai = {
+      endpoint = OPENAI_HOST,
+      secret = OPENAI_KEY,
+    },
+    groq = {
+      endpoint = GROQ_HOST,
+      secret = GROQ_KEY,
+    },
+  },
+
+  chat_shortcut_respond = { modes = { "n", "i", "v", "x" }, shortcut = "<C-g><cr>" },
+  chat_confirm_delete = false,
+
+  -- prefix for all commands
+  cmd_prefix = "Gp",
+
+  default_chat_agent = "GroqLLAMA_8B",
+  whisper = {
+    -- -- TODO: In the future, when gpnvim will support whisper options
+    endpoint = GROQ_AUDIO,
+    secret = GROQ_KEY,
+    model = GROQ_WHISPER_MODEL,
+    store_dir = "/tmp/gp_whisper"
+  },
+
+  agents = {
+    {
+      provider = "openai",
+      name = "ChatGPT4o",
+      chat = false,
+      command = true,
+      -- string with model name or table with model name and parameters
+      model = { model = "gpt-4o", temperature = 0.8, top_p = 1 },
+      -- system prompt (use this to specify the persona/role of the AI)
+      system_prompt = "You are an AI working as a code editor.\n\n"
+        .. "Please AVOID COMMENTARY OUTSIDE OF THE SNIPPET RESPONSE.\n"
+        .. "START AND END YOUR ANSWER WITH:\n\n```",
+    },
+    {
+      provider = "openai",
+      name = "ChatGPT4o-mini",
+      chat = true,
+      command = true,
+      -- string with model name or table with model name and parameters
+      model = { model = "gpt-4o-mini", temperature = 0.8, top_p = 1 },
+      -- system prompt (use this to specify the persona/role of the AI)
+      system_prompt = "You are an AI working as a code editor.\n\n"
+        .. "Please AVOID COMMENTARY OUTSIDE OF THE SNIPPET RESPONSE.\n"
+        .. "START AND END YOUR ANSWER WITH:\n\n```",
+    },
+    {
+      provider = "groq",
+      name = "GroqLLAMA_8B",
+      chat = true,
+      command = true,
+      -- string with model name or table with model name and parameters
+      model = { model = "llama-3.1-70b-versatile", temperature = 0.8, top_p = 1 },
+      system_prompt = "You are an AI helping the user with code and other tasks\n\n"
+        .. "Please AVOID COMMENTARY OUTSIDE OF THE SNIPPET RESPONSE.\n",
+    },
+    {
+      provider = "groq",
+      name = "GroqLLAMA_8B",
+      chat = true,
+      command = true,
+      -- string with model name or table with model name and parameters
+      model = { model = "llama-3.2-11b-text-preview", temperature = 0.8, top_p = 1 },
+      system_prompt = "Given a task or problem, please provide a concise and well-formatted solution or answer.\n\n"
+        .. "Please keep your response within a code snippet, and avoid unnecessary commentary.\n",
+    },
+  },
+}

lua/gp/config.lua

@@ -379,8 +379,12 @@ local config = {
 		-- you can disable whisper completely by whisper = {disable = true}
 		disable = false,
 
-		-- OpenAI audio/transcriptions api endpoint to transcribe audio to text
+		-- OpenAI compatible audio/transcriptions api endpoint to transcribe audio to text
 		endpoint = "https://api.openai.com/v1/audio/transcriptions",
+
+		-- Override the secret for whisper. It can be empty as well, in case of running local whisper server
+		secret = os.getenv("OPENAI_API_KEY"),
+
 		-- directory for storing whisper files
 		store_dir = (os.getenv("TMPDIR") or os.getenv("TEMP") or "/tmp") .. "/gp_whisper",
 		-- multiplier of RMS level dB for threshold used by sox to detect silence vs speech
@@ -404,6 +408,10 @@ local config = {
 		-- whisper_rec_cmd = {"arecord", "-c", "1", "-f", "S16_LE", "-r", "48000", "-d", "3600", "rec.wav"},
 		-- whisper_rec_cmd = {"ffmpeg", "-y", "-f", "avfoundation", "-i", ":0", "-t", "3600", "rec.wav"},
 		rec_cmd = nil,
+
+		-- Whisper model to use. See https://platform.openai.com/docs/models/whisper for more info for OpenAI models
+		-- Can also set a custom model. Locally, or groq-compatible models can be used.
+		model = "whisper-1",
 	},
 
 	-- image generation settings

lua/gp/whisper.lua

@@ -13,6 +13,7 @@ local vault = require("gp.vault")
 local default_config = require("gp.config")
 
 local W = {
+	--@class GpConfig
 	config = {},
 	cmd = {},
 	disabled = false,
@@ -52,11 +53,11 @@ local whisper = function(callback, language)
 		return
 	end
 
-	local bearer = vault.get_secret("openai_api_key")
-	if not bearer then
-		logger.error("OpenAI API key not found")
-		return
-	end
+	-- get the secret for whisper. It can be empty as well, in case of running local whisper server
+	local bearer = W.config.secret or vault.get_secret("openai_api_key")
+
+	-- get the model.
+	local model = W.config.model or "whisper-1"
 
 	local rec_file = W.config.store_dir .. "/rec.wav"
 	local rec_options = {
@@ -180,34 +181,46 @@ local whisper = function(callback, language)
 
 	-- transcribe the recording
 	local transcribe = function()
-		local cmd = "cd "
-			.. W.config.store_dir
-			.. " && "
-			.. "export LC_NUMERIC='C' && "
-			-- normalize volume to -3dB
-			.. "sox --norm=-3 rec.wav norm.wav && "
-			-- get RMS level dB * silence threshold
-			.. "t=$(sox 'norm.wav' -n channels 1 stats 2>&1 | grep 'RMS lev dB' "
+		local cd_cmd = "cd " .. W.config.store_dir
+		local export_lc_numeric_cmd = "export LC_NUMERIC='C'"
+		local sox_norm_cmd = "sox --norm=-3 rec.wav norm.wav"
+		local sox_silence_t = "t=$(sox 'norm.wav' -n channels 1 stats 2>&1 | grep 'RMS lev dB' "
 			.. " | sed -e 's/.* //' | awk '{print $1*"
 			.. W.config.silence
-			.. "}') && "
-			-- remove silence, speed up, pad and convert to mp3
-			.. "sox -q norm.wav -C 196.5 final.mp3 silence -l 1 0.05 $t'dB' -1 1.0 $t'dB'"
-			.. " pad 0.1 0.1 tempo "
+			.. "}')"
+		local remove_silence_cmd = "sox -q norm.wav -C 196.5 final.mp3 silence -l 1 0.05 $t'dB' -1 1.0 $t'dB' pad 0.1 0.1 tempo "
 			.. W.config.tempo
-			.. " && "
-			-- call openai
-			.. curl
+
+		local curl_bearer_header = ""
+		if bearer ~= "" then
+			curl_bearer_header = '-H "Authorization: Bearer ' .. bearer .. '" '
+		end
+
+		local curl_cmd = curl
 			.. " --max-time 20 "
 			.. W.config.endpoint
-			.. ' -s -H "Authorization: Bearer '
-			.. bearer
-			.. '" -H "Content-Type: multipart/form-data" '
-			.. '-F model="whisper-1" -F language="'
+			.. " -s "
+			.. curl_bearer_header
+			.. '-H "Content-Type: multipart/form-data" '
+			.. '-F model="'
+			.. model
+			.. '" -F "language='
 			.. language
-			.. '" -F file="@final.mp3" '
+			.. '" -F "[email protected]" '
 			.. '-F response_format="json"'
 
+		local cmd = cd_cmd
+			.. " && "
+			.. export_lc_numeric_cmd
+			.. " && "
+			.. sox_norm_cmd
+			.. " && "
+			.. sox_silence_t
+			.. " && "
+			.. remove_silence_cmd
+			.. " && "
+			.. curl_cmd
+
 		tasker.run(nil, "bash", { "-c", cmd }, function(code, signal, stdout, _)
 			if code ~= 0 then
 				logger.error(string.format("Whisper query exited: %d, %d", code, signal))