add support for MammothModa2 model

examples/offline_inference/mammothmodal2_preview/mammoth_moda2_image_summary.yaml

@@ -0,0 +1,18 @@
+stage_args:
+  - stage_id: 0
+    runtime:
+      devices: "0"
+      max_batch_size: 16
+    engine_args:
+      model_stage: ar
+      model_arch: MammothModa2ForConditionalGeneration
+      worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker
+      scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
+      max_model_len: 8192
+      gpu_memory_utilization: 0.5
+      enforce_eager: true
+      trust_remote_code: true
+      engine_output_type: text
+      enable_prefix_caching: false
+    final_output: true
+    final_output_type: text

examples/offline_inference/mammothmodal2_preview/run_mammothmoda2_image_summary.py

@@ -0,0 +1,147 @@
+"""
+Offline inference example: MammothModa2 image summarization (single AR stage).
+
+Example:
+  uv run python examples/offline_inference/mammothmodal2_preview/run_mammothmoda2_image_summary.py \
+    --model /data/datasets/models-hf/MammothModa2-Preview \
+    --image /path/to/input.jpg \
+    --question "Please summarize the content of this image."
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+from pathlib import Path
+
+from PIL import Image
+from vllm import SamplingParams
+from vllm.multimodal.image import convert_image_mode
+
+from vllm_omni import Omni
+
+DEFAULT_SYSTEM = "You are a helpful assistant."
+DEFAULT_QUESTION = "Please summarize the content of this image."
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="MammothModa2 image summarization (offline, AR only).")
+    parser.add_argument(
+        "--model",
+        type=str,
+        default="/data/datasets/models-hf/MammothModa2-Preview",
+        help="Path to model directory or model id.",
+    )
+    parser.add_argument(
+        "--stage-config",
+        type=str,
+        default=str(Path(__file__).with_name("mammoth_moda2_image_summary.yaml")),
+        help="Path to stage config yaml (single-stage AR->text).",
+    )
+    parser.add_argument(
+        "--image",
+        type=str,
+        required=True,
+        help="Path to input image.",
+    )
+    parser.add_argument(
+        "--question",
+        type=str,
+        default=DEFAULT_QUESTION,
+        help="Question/instruction for the model.",
+    )
+    parser.add_argument(
+        "--system",
+        type=str,
+        default=DEFAULT_SYSTEM,
+        help="System prompt.",
+    )
+    parser.add_argument(
+        "--max-tokens",
+        type=int,
+        default=512,
+        help="Max new tokens to generate.",
+    )
+    parser.add_argument("--temperature", type=float, default=0.2)
+    parser.add_argument("--top-p", type=float, default=0.9)
+    parser.add_argument("--seed", type=int, default=42)
+    parser.add_argument("--trust-remote-code", action="store_true")
+    parser.add_argument(
+        "--out",
+        type=str,
+        default="image_summary.txt",
+        help="Path to save output text.",
+    )
+    return parser.parse_args()
+
+
+def build_prompt(system: str, question: str) -> str:
+    return (
+        f"<|im_start|>system\n{system}<|im_end|>\n"
+        "<|im_start|>user\n"
+        "<|vision_start|><|image_pad|><|vision_end|>"
+        f"{question}<|im_end|>\n"
+        "<|im_start|>assistant\n"
+    )
+
+
+def main() -> None:
+    args = parse_args()
+
+    if not os.path.exists(args.image):
+        raise FileNotFoundError(f"Image file not found: {args.image}")
+
+    os.makedirs(os.path.dirname(args.out) or ".", exist_ok=True)
+
+    pil_image = Image.open(args.image)
+    image_data = convert_image_mode(pil_image, "RGB")
+    prompt = build_prompt(args.system, args.question)
+
+    omni = Omni(
+        model=args.model,
+        stage_configs_path=args.stage_config,
+        trust_remote_code=args.trust_remote_code,
+    )
+    try:
+        sp = SamplingParams(
+            temperature=float(args.temperature),
+            top_p=float(args.top_p),
+            top_k=-1,
+            max_tokens=int(args.max_tokens),
+            seed=int(args.seed),
+            detokenize=True,
+        )
+        outputs = omni.generate(
+            [
+                {
+                    "prompt": prompt,
+                    "multi_modal_data": {"image": image_data},
+                }
+            ],
+            [sp],
+        )
+    finally:
+        omni.close()
+
+    if not isinstance(outputs, list):
+        outputs = [outputs]
+
+    lines: list[str] = []
+    for stage_outputs in outputs:
+        req_outputs = getattr(stage_outputs, "request_output", stage_outputs)
+        req_outputs = req_outputs if isinstance(req_outputs, list) else [req_outputs]
+        for ro in req_outputs:
+            text = ro.outputs[0].text if getattr(ro, "outputs", None) else str(ro)
+            lines.append(f"request_id: {getattr(ro, 'request_id', 'unknown')}\n")
+            lines.append("answer:\n")
+            lines.append(text.strip() + "\n")
+            lines.append("\n")
+
+    with open(args.out, "w", encoding="utf-8") as f:
+        f.writelines(lines)
+
+    print(f"[OK] Saved summary to: {args.out}")
+
+
+if __name__ == "__main__":
+    main()