[model validation]: olmo-3-1025-7B

docs/models/samples/Olmo-3-1025-7B.py

@@ -0,0 +1,58 @@
+# -*- coding: utf-8 -*-
+"""
+Requirements:
+- vLLM: v0.11.0 or higher
+- vLLM-metax: v0.11.0 or higher
+- MACA SDK: 3.2.x.x or higher
+"""
+
+import argparse
+import sys
+import torchvision
+torchvision.disable_beta_transforms_warning()
+
+from vllm import LLM, SamplingParams
+
+PROMPTS = [
+    "Describe vLLM's core advantages in one paragraph.",
+    "Give me 3 engineering recommendations to improve large language model throughput, and briefly explain why.",
+    "Explain what tensor parallelism is and when to use it.",
+    "Translate this sentence into Chinese: We are using vLLM to perform offline inference validation testing.",
+    "Write a Python snippet (no more than 5 lines) showing how to use vLLM to call a local model to generate text.",
+]
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default="allenai/Olmo-3-1025-7B", help="Model name or path.")
+    parser.add_argument("--max-tokens", type=int, default=128, help="Max tokens.")
+    parser.add_argument("--temperature", type=float, default=0.2, help="Temp.")
+    args = parser.parse_args()
+
+    sampling = SamplingParams(
+        temperature=args.temperature,
+        max_tokens=args.max_tokens,
+        n=1,
+    )
+
+    print(f"[INFO] Loading model: {args.model}")
+    llm = LLM(
+        model=args.model,
+        dtype="auto",
+        # WARNING: `trust_remote_code=True` is a security risk as it allows for arbitrary code execution from the model's repository. Only use this with models you trust.
+        trust_remote_code=True,
+        tensor_parallel_size=1,
+        gpu_memory_utilization=0.90,
+    )
+
+    print(f"[INFO] Generating {len(PROMPTS)} completions")
+    outputs = llm.generate(PROMPTS, sampling)
+
+    for i, (prompt, out) in enumerate(zip(PROMPTS, outputs), 1):
+        text = out.outputs[0].text.strip() if out.outputs else ""
+        print(f"\n[{i}]:\nUser: <<< {prompt}\nAssistant: >>> {text}")
+
+    success = any(o.outputs and o.outputs[0].text.strip() for o in outputs)
+    sys.exit(0 if success else 2)
+
+if __name__ == "__main__":
+    main()

docs/models/supported_models.md

@@ -45,6 +45,7 @@ Here the plugin would list all the **tested** model on Maca.
 | `Qwen3ForCausalLM` | Qwen3 | `Qwen/Qwen3-8B`, etc. | ✅︎ | ✅︎ |
 | `Qwen3MoeForCausalLM` | Qwen3MoE | `Qwen/Qwen3-30B-A3B`, etc. | ✅︎ | ✅︎ |
 | `Qwen3NextForCausalLM` | Qwen3NextMoE | `Qwen/Qwen3-Next-80B-A3B-Instruct`, etc. | ✅︎ | ✅︎ |
+| `OLMo3ForCausalLM` | OLMo3 | `TBA, allenai/Olmo-3-1025-7B`, etc. | ✅︎ | ✅︎ |
 
 ## List of Multimodal Language Models