Add model support for granite speech plus (#43519)

Signed-off-by: Zvi Kons[WSL] <zvi@il.ibm.com>
Signed-off-by: Zvi Kons (BlueVela) <zvi@il.ibm.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
Zvi Kons
2026-06-04 14:47:48 +00:00
committed by GitHub
co-authored by Claude Opus 4.7
parent 06ee2d8433
commit b21443e23c
6 changed files with 106 additions and 3 deletions
@@ -30,11 +30,14 @@ def vllm_to_hf_output(
MODEL_NAME = "ibm-granite/granite-speech-3.3-2b"
MODEL_NAME_4_0 = "ibm-granite/granite-4.0-1b-speech"
# "plus" variant of granite speech (uses GraniteSpeechPlusForConditionalGeneration).
MODEL_NAME_4_1_PLUS = "ibm-granite/granite-speech-4.1-2b-plus"
# Audio lora co-exists directly in the 3.3 model directory,
# the 4.0 model has adapters merged into the weights.
# the 4.0 and 4.1-plus models have adapters merged into the weights.
models: dict[str, str | None] = {
MODEL_NAME: MODEL_NAME,
MODEL_NAME_4_0: None,
MODEL_NAME_4_1_PLUS: None,
}
+4
View File
@@ -938,6 +938,10 @@ _MULTIMODAL_EXAMPLE_MODELS = {
"ibm-granite/granite-speech-3.3-2b",
extras={"4.0-1b": "ibm-granite/granite-4.0-1b-speech"},
),
"GraniteSpeechPlusForConditionalGeneration": _HfExamplesInfo(
"ibm-granite/granite-speech-4.1-2b-plus",
min_transformers_version="5.8.0",
),
"GLM4VForCausalLM": _HfExamplesInfo(
"zai-org/glm-4v-9b",
trust_remote_code=True,