mirror of
https://github.com/ollama/ollama.git
synced 2026-11-12 03:46:14 -06:00
Daniel HiltgenandGitHub
3fbc609be4
Fixes for running LFM3 GGUFs on the llama-server runner, plus GGUF text-to-speech that turns on once llama.cpp is bumped past b11901. - BOS: honor an explicit tokenizer.ggml.add_bos_token. LFM3 GGUFs (and lfm2:24b) set it to false and expect the template's BOS, which the lfm2 default was stripping. The lfm2 default still applies when the key is absent. On lfm2:24b, WikiText PPL goes from 81.3 to 72.0. - repeat_last_n -1 now maps to the context size; llama-server rejects negative values. - Audio-only projectors (LFM3-Audio) no longer advertise vision, so clients don't send images the model can't take. - /v1/audio/speech on GGUF models calls llama-server /tts chunk by chunk. Renderers opt in through a new SpeechPrompter (voices, voice parsing, chunking), so other speech models can add their own. The LFM3 voice and chunking helpers move from the MLX runner into model/renderers and are shared by both runners.
36 lines
1.3 KiB
Go
36 lines
1.3 KiB
Go
package renderers
|
|
|
|
// SpeechVoice is the voice and language a llama-server /tts request asks for.
|
|
type SpeechVoice struct{ Voice, Language string }
|
|
|
|
// SpeechChunk is one /tts generation of a speech request.
|
|
type SpeechChunk struct {
|
|
// Text is read aloud, escaped for the model's tts template.
|
|
Text string
|
|
// MaxFrames caps the codec frames the generation may produce.
|
|
MaxFrames int
|
|
}
|
|
|
|
// SpeechPrompter maps speech requests onto llama-server's /tts for the models
|
|
// a renderer serves; llama-server renders the model's own TTS template.
|
|
type SpeechPrompter interface {
|
|
// SpeechVoices lists the voice names to advertise; renderers whose models
|
|
// do not speak return none.
|
|
SpeechVoices() []string
|
|
// SpeechVoice resolves a voice name; the client gets a 400 on error.
|
|
SpeechVoice(name string) (SpeechVoice, error)
|
|
// SpeechChunks splits input into generations, or returns nil when there is
|
|
// no text to speak; count returns the token count of a piece of text.
|
|
SpeechChunks(input string, count func(string) int) []SpeechChunk
|
|
}
|
|
|
|
// SpeechPrompterForRenderer returns the named renderer's SpeechPrompter if
|
|
// its models read text aloud through llama-server.
|
|
func SpeechPrompterForRenderer(name string) (SpeechPrompter, bool) {
|
|
p, ok := rendererForName(name).(SpeechPrompter)
|
|
if !ok || len(p.SpeechVoices()) == 0 {
|
|
return nil, false
|
|
}
|
|
return p, true
|
|
}
|