Ornith-1.5-122B — NVFP4 (mixed) for veloGB10

Original model: ornith-ai/Ornith-1.5-122B by ornith-ai. This repository holds quantized (modified) weights of it; the original model card follows this one, verbatim.

What this pack is

Quantized weights: produced by us from ornith-ai/Ornith-1.5-122B with veloGB10's offline quantizer, recipe all,gdn:fp8 (NVFP4 GEMMs, FP8 Gated DeltaNet projections; MTP head included); compressed-tensors layout, nvfp4-pack-quantized format.
Unchanged from the original: tokenizer, chat template and configuration (apart from the quantization_config block).
Original model: ornith-ai/Ornith-1.5-122B.
Architecture: the Qwen3.5-122B-A10B layout (qwen3_5_moe); ple_ngram_nvfp4.bin is an engine-side table other frameworks ignore.

Built for veloGB10

This pack is published primarily for the veloGB10 inference engine (gb10_inference) on one, two or four NVIDIA DGX Spark / GB10 machines: it is the pack we load, test and benchmark there, and the commands below are the ones it was tested with. The weights are standard compressed-tensors format, so any framework that reads the format can use them too.

Running it with veloGB10

Download a release binary of veloGB10 (one self-contained file; see its README), and the pack with hf download doth4580/Ornith-1.5-122B-NVFP4-mixed --local-dir /path/to/Ornith-1.5-122B-NVFP4-mixed. The server is OpenAI-compatible on http://<head-ip>:9000/v1. The flags below are the ones this pack was tested with; --max-batch sets how many requests run at once.

One GB10 (TP=1)

./gb10_inference --server \
  --model-dir /path/to/Ornith-1.5-122B-NVFP4-mixed \
  --host 0.0.0.0 --port 9000 \
  --max-seq-len 262144 \
  --max-batch 2 \
  --max-tokens 65536 \
  --default-presence-penalty 1.5 \
  --prefix-cache on \
  --mtp=auto

Two GB10s (TP=2)

On the second machine (no model copy, no configuration):

./gb10_inference --node --port 29500

On the head (the machine with the model folder):

./gb10_inference --server \
  --model-dir /path/to/Ornith-1.5-122B-NVFP4-mixed \
  --tp 2 --nodes <node-ip>:29500 \
  --host 0.0.0.0 --port 9000 \
  --max-seq-len 262144 \
  --max-batch 2 \
  --max-tokens 65536 \
  --default-presence-penalty 1.5 \
  --prefix-cache on \
  --mtp=auto

Four GB10s (TP=4)

Start ./gb10_inference --node --port 29500 on each of the three other machines, then on the head:

./gb10_inference --server \
  --model-dir /path/to/Ornith-1.5-122B-NVFP4-mixed \
  --tp 4 --nodes <node1-ip>:29500,<node2-ip>:29500,<node3-ip>:29500 \
  --host 0.0.0.0 --port 9000 \
  --max-seq-len 262144 \
  --max-batch 2 \
  --max-tokens 65536 \
  --default-presence-penalty 1.5 \
  --prefix-cache on \
  --mtp=auto

The nodes need no copy of the model. The head sends each node only its own share of the weights over the network and keeps it in a local cache, so a second start of the same model transfers nothing. On a box with two active RoCE port pairs, name the one to use with --rdma-dev <rail1>,<rail2>; with one pair it is found automatically.

With the DFlash2 drafter for Ornith-1.5-122B (incoai/Ornith-1.5-122B-DFlash2): add --spec-source dflash2 --draft-dir /path/to/Ornith-1.5-122B-DFlash2 --df2-block 16 to the head's command line (any topology above).

Measured with veloGB10

VeloBench runs against the served API on NVIDIA DGX Spark-class GB10 machines (2026-11-09, veloGB10 0.7.7 development builds), at production sampling: each request sends the pack's default sampler settings, thinking on, one request at a time for the single-request columns. Decode is in output tokens per second and includes speculative decoding. One run per cell; sampled runs vary by roughly ±10%.

Topology	Draft source	Code turn decode	Story turn decode	All-turn median decode	Prefill (tok/s)	2 requests, aggregate	Draft acceptance
TP=1	pack's MTP head	29	24	26	695	39	44.7%
TP=2	pack's MTP head	41	37	37	1,059	59	34.3%
TP=4	pack's MTP head	44	39	41	997	69	34.0%
TP=1	Ornith DFlash2	32	9	12	678	40	59.2%
TP=2	Ornith DFlash2	56	13	18	1,045	59	55.4%
TP=4	Ornith DFlash2	58	17	21	985	70	57.7%

Intellectual property

We do not claim any intellectual-property ownership of the model, its weights, its architecture or its original documentation. All rights stay with the original authors and with every contributor credited above, under their own licences, which apply to this repository unchanged. Our contribution is limited to what the "What this pack is" section says we did (quantization, packaging, testing and the veloGB10 instructions). If you are a rights holder and want something here changed or removed, open a discussion on this repository.

Original model card — ornith-ai/Ornith-1.5-122B

Reproduced verbatim from the original repository (metadata header removed).

Ornith-1.5-122B

Chirp Chirp! 🐦 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.

Ornith-1.5 extends Ornith-1.0 (which was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training) by expanding the self-improvement loop from scaffold and rollout optimization to jointly optimizing task generation, scaffold construction, and solution rollouts. Rather than relying on a fixed set of human-curated tasks and manually designed harnesses, Ornith-1.5 continuously generates new training tasks, discovers effective strategies for solving them, and improves the policy through reinforcement learning. For more details on the task, harness, and rollout reward design, please refer to our blog. Ornith 1.5 122B Benchmark Results

Ornith 1.5 122B

This model card documents Ornith-1.5-122B, the most capable and efficient.member of the Ornith-1.5 family — a 122B MOE model designed for efficient single-GPU deployment, and edge-deployable on mobile devices via its quantized Ornith-1.5-122B-Mobile variant.

Benchmarks

	Ornith-1.5-122B	Qwen3.8-Flash-Next _(125B)	Qwen3.8 _{(27B dense)}	Nemotron 3.5 Super VL _(120B)	DeepSeek-V4-Flash-0731 _(284B)	Beam _(501B)
Coding
TerminalBench 2.1	81.200	83.400	71.900	66.700	82.700	80.100
CyberGym	80.000	78.000	55.000	34.000	76.700	59.000
NL2Repo	51.200	45.200	42.300	21.000	54.200	49.400
SWE-bench Pro _(no-hack)	76.900	43.900	65.500	42.500	-	57.000
SWE-bench Verified _(non-hack)	84.500	82.800	76.800	61.200	-	-
SWE-bench Multilingual _(non-hack)	76.400	73.000	70.300	48.000	-	-
Reasoning
HLE _{(with tools)}	48.000	38.000	31.300	33.200	50.800	51.100
HLE _{(w/o tools)}	37.680	34.160	29.320	31.790	-	-
SciCode	53.800	47.500	43.600	47.100	-	-
Agentic
WideSearch	75.200	71.800	62.600	69.200	77.300	68.500
BrowseComp	78.500	75.300	71.400	61.000	80.200	67.600
MCP-Atlas	74.600	74.100	69.000	63.300	74.600	-
Toolathlon-Verified	70.800	71.400	67.100	60.100	70.300	63.500
ClawEval	79.400	76.500	75.400	61.200	-	-

* Terminal-Bench 2.1 (Claude Code): We evaluate Terminal-Bench 2.1 using Claude Code 2.1.126 with parser=json, temperature=1.0, top_p=1.0, max_new_tokens=131072. Results are averaged over 5 runs. Again, Qwen chat template needs to be modified.
* SWE-Bench Verified, Pro, and Multilingual: Evaluated using the OpenHands harness with temperature=1.0, top_p=0.95, and a 256K context window. Anti-hacking safeguards are applied throughout evaluation: models are explicitly informed of potential reward-hacking behaviors and instructed to avoid them. Any solution that still exhibits such behavior is marked incorrect.
* NL2Repo: with temperature=1.0, top_p=1.0, 400K context, 48K output. Access to specified GitHub repositories and pip packages is blocked to prevent reward hacking.
* HLE: Evaluated using Claude 4.6 Opus as the judge model.
* MCP-Atlas: All models were evaluated in thinking mode on the 500-task public subset, with a 10-minute timeout per task. We use Claude 4.8 Opus as the judge model.
* Toolathlon-Verified: We use the official evaluation service with the maximum token limit set to 128K.
* ClawEval: An agentic code benchmark over real-user task distributions; temp=0.6 and 256K context.

Quickstart

📝 NOTE

Ornith-1.5-122B is a reasoning model: by default the assistant turn opens with a <think> … </think> block before the final answer. The serving recipes below enable a reasoning parser so the chain-of-thought is returned in a separate reasoning_content field, and a tool-call parser so the model's <tool_call> blocks are surfaced as OpenAI-style tool_calls.

Serving Ornith-1.5-122B requires recent runtimes:

Transformers ≥ 5.8.1
vLLM ≥ 0.19.1
SGLang ≥ 0.5.9

Recommended sampling parameters:

For general tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
For precise coding tasks: temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0

Serving Ornith-1.5-122B

Ornith-1.5-122B is a MOE 122B model, so it serves on a single 80GB GPU. The recipes below stand up an OpenAI-compatible server; add --tensor-parallel-size / --tp if you want to shard across more GPUs.

vLLM

vllm serve ornith-ai/Ornith-1.5-122B --served-model-name Ornith-1.5-122B --host 0.0.0.0 --port 8000 --max-model-len 262144 --gpu-memory-utilization 0.90 --enable-prefix-caching --enable-auto-tool-choice --tool-call-parser qwen3_xml --reasoning-parser qwen3 --trust-remote-code

SGLang

python -m sglang.launch_server --model-path ornith-ai/Ornith-1.5-122B --served-model-name Ornith-1.5-122B --host 0.0.0.0 --port 8000 --context-length 262144 --mem-fraction-static 0.85 --tool-call-parser qwen3_coder --reasoning-parser qwen3

For Long-Context

Ornith-1.5-122B handles context windows of up to 262,144 tokens. When a task's combined input and output must go beyond this limit, we suggest extending the effective window with RoPE scaling — YaRN is the technique we validate against, and it is already built into both vLLM and SGLang. With a scaling factor of 4.0, the usable window grows to roughly 1M tokens.

You can turn YaRN on in either of two ways:

Edit the checkpoint's config.json. Add a rope_scaling block to the model configuration:

{
    "rope_scaling": {
        "rope_type": "yarn",
        "factor": 4.0,
        "original_max_position_embeddings": 262144
    }
}

Override at launch time. Leave the checkpoint untouched and extend the serve commands above with the equivalent flags.

vLLM:

VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ornith-ai/Ornith-1.5-122B ... --hf-overrides '{"rope_scaling": {"rope_type": "yarn", "factor": 4.0, "original_max_position_embeddings": 262144}}' --max-model-len 1000000

SGLang:

SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"rope_scaling": {"rope_type": "yarn", "factor": 4.0, "original_max_position_embeddings": 262144}}' --context-length 1000000

📝 NOTE

Open-source runtimes implement YaRN statically: the same scaling factor is applied to every request regardless of its length, which can slightly hurt quality on ordinary-length inputs. Only enable rope_scaling when your workload genuinely needs the longer window, and size factor to match it — the target window is roughly factor × 262,144, so if your requests top out around 524,288 tokens, factor: 2.0 is the better setting.

Using Ornith-1.5-122B via the Chat Completions API

Once a vLLM or SGLang server is running, talk to it with any OpenAI-compatible client.

Basic Usage

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="EMPTY",  # any non-empty string works for a local server
)

response = client.chat.completions.create(
    model="Ornith-1.5-122B",
    messages=[
        {"role": "user", "content": "Write a one-line Python lambda that squares a number."}
    ],
    temperature=0.6,
    top_p=0.95,
    max_tokens=1024,
)

message = response.choices[0].message
# reasoning_content holds the <think> trace; content holds the final answer.
print("reasoning:", getattr(message, "reasoning_content", None))
print("answer:", message.content)

You can also stream tokens, or hand the model tools — Ornith-1.5-122B emits well-formed function calls that the server parses into the standard tool_calls field:

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get the current weather for a city",
            "parameters": {
                "type": "object",
                "properties": {"city": {"type": "string"}},
                "required": ["city"],
            },
        },
    }
]

response = client.chat.completions.create(
    model="Ornith-1.5-122B",
    messages=[{"role": "user", "content": "What is the weather in Paris right now?"}],
    tools=tools,
    tool_choice="auto",
    temperature=0.6,
    max_tokens=2048,
)

tool_call = response.choices[0].message.tool_calls[0]
print(tool_call.function.name, tool_call.function.arguments)
# -> get_weather {"city": "Paris"}

You can point any OpenAI-compatible SDK (Python, Node.js, etc.) or curl at the same /v1/chat/completions endpoint.

Agentic Usage

Ornith-1.5-122B exposes an OpenAI-compatible endpoint with tool calling, it works out of the box with standard agent frameworks.

Examples of using Ornith with agents:

Ollama

ollama run ornith-1.5:122b

Atomic.chat

# Both runtimes load a GGUF build of Ornith (publish one at ornith-ai/Ornith-1.5-122B-GGUF).

# llama.cpp — serve an OpenAI-compatible API on port 8000.
llama-server -hf hf.co/ornith-ai/Ornith-1.5-122B-GGUF --port 8000 -c 262144

llama.cpp

# Both runtimes load a GGUF build of Ornith (publish one at ornith-ai/Ornith-1.5-122B-GGUF).

# llama.cpp — serve an OpenAI-compatible API on port 8000.
llama-server -hf hf.co/ornith-ai/Ornith-1.5-122B-GGUF --port 8000 -c 262144

Hermes Agent

# Hermes talks to any OpenAI-compatible endpoint — point it at your Ornith server.
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"
export MODEL="ornith-ai/Ornith-1.5-122B"

OpenClaw

# OpenClaw talks to any OpenAI-compatible endpoint — point it at your Ornith server.
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"
export OPENAI_MODEL="ornith-ai/Ornith-1.5-122B"

Unsloth Studio

pip install unsloth

# Load Ornith for fast local inference or fine-tuning (Python):
#   from unsloth import FastLanguageModel
#   model, tokenizer = FastLanguageModel.from_pretrained(
#       "unsloth/Ornith-1.5-122B-GGUF",
#       max_seq_length=262144,
#       load_in_4bit=True,
#   )

Coding CLIs

Ornith-1.5-122B is optimized for terminal-based coding agents. Point any OpenAI-compatible coding CLI at your Ornith-1.5-122B endpoint (set OPENAI_BASE_URL and OPENAI_API_KEY) to understand large codebases, automate tedious work, and ship faster.

OpenCode

# Register your local Ornith endpoint as a provider in ~/.config/opencode/opencode.json:
#
# {
#   "$schema": "https://opencode.ai/config.json",
#   "provider": {
#     "ornith": {
#       "npm": "@ai-sdk/openai-compatible",
#       "name": "Ornith (local)",
#       "options": { "baseURL": "http://localhost:8000/v1", "apiKey": "EMPTY" },
#       "models": { "ornith-ai/Ornith-1.5-122B": { "name": "Ornith-1.5-122B" } }
#     }
#   }
# }

opencode

Citation

If you find our work helpful, feel free to give us a cite.

@misc{ornith_1_5,
    title = {{Ornith-1.5}: From Self-Scaffolding to Self-Improvement},
    url = {https://ornith.ai/ornith_1_5.html},
    author = {{Ornith Team}},
    year = {2026}
}

Downloads last month: 21

Safetensors

Model size

125B params

Tensor type

BF16

F8_E4M3

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for doth4580/Ornith-1.5-122B-NVFP4-mixed

Base model

ornith-ai/Ornith-1.5-122B

Quantized

(55)

this model