Clarify Preview checkpoint identity and benchmark scope

README.md

@@ -12,40 +12,53 @@ tags:
 - long-context
 - hybrid-attention
 ---
-
 <p align="center">
   <img width="132" src="assets/agnes_logo.svg" alt="Agnes AI logo">
 </p>
 
 <p align="center">
   <a href="https://agnes-ai.com/"><img src="https://img.shields.io/badge/Agnes_AI-Website-3248AF" alt="Agnes AI website"></a>
-  <a href="#quickstart"><img src="https://img.shields.io/badge/Agnes--3.0--Flash-Open_Weights-3248AF" alt="Open weights"></a>
+  <a href="#quickstart"><img src="https://img.shields.io/badge/Agnes--3.0--Flash_Preview-Open_Weights-3248AF" alt="Open weights"></a>
   <img src="https://img.shields.io/badge/License-Apache_2.0-111827" alt="Apache 2.0">
 </p>
 
-# Agnes-3.0-Flash
 
-Hello! 👋 Today we are introducing **Agnes-3.0-Flash**, an **open-weights multimodal model** built for people who want flagship-class reasoning without flagship-class hardware.
+# Agnes-3.0-Flash Preview
 
-Highlights:
 
-- **Competitive across core capabilities.** Agnes-3.0-Flash posts competitive results across reasoning, coding, and instruction-following evaluations.
+## Model version clarification
+
+This repository contains an earlier open-weight **Preview checkpoint** of Agnes 3.0 Flash. It is distinct from the newer **production/API checkpoint** listed on [Artificial Analysis](https://artificialanalysis.ai/models/agnes-3-0-flash).
+
+The Preview release has **33B parameters** and a context window of **262,144 tokens**. The production/API model uses a different checkpoint and configuration, with a **1M-token context window**. Its benchmark results should not be attributed to the Preview weights released here.
+
+This repository was initially published as `Agnes-3.0-Flash` without the `Preview` suffix. The model card now explicitly identifies this release as **Agnes-3.0-Flash Preview** to clarify the distinction between the open-weight release and the production/API model.
+
+The specifications and Agnes benchmark results below refer to the **Preview checkpoint**.
+
+Hello! 👋 Today we are introducing **Agnes-3.0-Flash Preview**, an **open-weights multimodal preview model** built for people who want flagship-class reasoning without flagship-class hardware.
+Highlights:
+- **Competitive across core capabilities.** Agnes-3.0-Flash Preview posts competitive results across reasoning, coding, and instruction-following evaluations.
 - **Built for demanding work.** A **262 144-token context window**, adjustable reasoning effort, tool calling, and **text, image and video** understanding.
 
+
+### Benchmarks
+
+> **Benchmark scope:** The Agnes results in the chart and table below belong to the **Agnes-3.0-Flash Preview open-weight checkpoint** released in this repository. They are not results for the production/API Agnes 3.0 Flash model listed on Artificial Analysis.
+
 <p align="center">
-  <img style="width:100%;max-width:1100px" src="assets/agnes_benchmarks.svg" alt="Agnes-3.0-Flash benchmark reference results">
+  <img style="width:100%;max-width:1100px" src="assets/agnes_benchmarks.svg" alt="Agnes-3.0-Flash Preview benchmark reference results">
 </p>
-## Agnes-3.0-Flash
 
-### Benchmarks
 
-Reference results across contemporary models are shown below. The figures were compiled from different sources, harnesses, and model snapshots and do not constitute a controlled head-to-head comparison.
+The label "Agnes-3.0-Flash" in the benchmark image refers to this Preview checkpoint.
 
+Reference results across contemporary models are shown below. The figures were compiled from different sources, harnesses, and model snapshots and do not constitute a controlled head-to-head comparison.
 <div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;width:100%;margin:0 auto;padding:16px 0;overflow-x:auto">
 <table style="width:100%;table-layout:fixed;border-collapse:collapse;font-size:11px;min-width:1180px">
 <thead><tr>
 <th style="width:15%;padding:9px 6px;text-align:left;border-bottom:2px solid #3248AF;color:#3248AF;font-size:11px">Benchmark</th>
-<th style="width:10%;padding:9px 5px;text-align:center;font-weight:700;border-bottom:2px solid #3248AF;color:#3248AF;background:rgba(50,72,175,.09);font-size:10px;line-height:1.3">Agnes-3.0-Flash</th>
+<th style="width:10%;padding:9px 5px;text-align:center;font-weight:700;border-bottom:2px solid #3248AF;color:#3248AF;background:rgba(50,72,175,.09);font-size:10px;line-height:1.3">Agnes-3.0-Flash Preview</th>
 <th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Qwen3.6-35B-A3B<br><span style="opacity:.65">35B / 3B active</span></th>
 <th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Kimi K2.5<br><span style="opacity:.65">1T / 32B active</span></th>
 <th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Muse Glimmer<br><span style="opacity:.65">30B</span></th>
@@ -67,10 +80,9 @@ Reference results across contemporary models are shown below. The figures were c
 Higher is better for every row. Header parameter figures mix total and active counts, and harnesses and snapshot dates differ across sources, so treat cross-column comparisons as reference values rather than a controlled head-to-head evaluation.
 </p>
 
-## Architecture
-
-Agnes-3.0-Flash is a hybrid-attention decoder: three of every four layers run a gated delta rule (recurrent, with per-layer state independent of sequence length), and the fourth runs standard global attention. Only 18 of the 72 layers therefore hold a KV cache that grows with context.
 
+## Architecture
+Agnes-3.0-Flash Preview is a hybrid-attention decoder: three of every four layers run a gated delta rule (recurrent, with per-layer state independent of sequence length), and the fourth runs standard global attention. Only 18 of the 72 layers therefore hold a KV cache that grows with context.
 | | |
 |---|---|
 | Context length | **262 144** tokens |
@@ -84,31 +96,28 @@ Agnes-3.0-Flash is a hybrid-attention decoder: three of every four layers run a
 | Vision tower | 27 layers, hidden 1152, patch 16, 2 × 2 spatial merge, projected to 5120 |
 
 ## Quickstart
-
 <div style="border-left:4px solid #3248AF;background:rgba(50,72,175,.08);border-radius:6px;padding:12px 16px;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;font-size:14px;line-height:1.6">
 <div style="font-weight:700;color:#3248AF;margin-bottom:6px">REMOTE CODE REQUIRED</div>
-<p style="margin:0"><b>Agnes-3.0-Flash</b> ships its own model implementation. Always load it with <code>trust_remote_code=True</code>.</p>
+
+<p style="margin:0"><b>Agnes-3.0-Flash Preview</b> ships its own model implementation. Always load it with <code>trust_remote_code=True</code>.</p>
+
 </div>
 
-### Requirements
 
+### Requirements
 ```bash
 pip install "transformers>=5.12" torch torchvision accelerate
 ```
-
 Tested on transformers 5.12.1. Image and video inputs go through the bundled processor, which needs `torchvision`.
 
 ### Transformers
-
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
-
 path = "Agnes-AI/Agnes-3.0-Flash"
 tok = AutoTokenizer.from_pretrained(path)
 model = AutoModelForCausalLM.from_pretrained(
     path, dtype="bfloat16", device_map="auto", trust_remote_code=True
 )
-
 msgs = [{"role": "user", "content": "请用三句话解释什么是人工智能。"}]
 ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)
 out = model.generate(ids, max_new_tokens=256)
@@ -116,12 +125,9 @@ print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))
 ```
 
 ### Images and video
-
 Image and video inputs go through the bundled processor (also remote code):
-
 ```python
 from transformers import AutoProcessor
-
 proc = AutoProcessor.from_pretrained(path, trust_remote_code=True)
 msgs = [{"role": "user", "content": [{"type": "image", "image": "photo.jpg"},
                                      {"type": "text", "text": "描述这张图。"}]}]
@@ -132,18 +138,14 @@ print(proc.batch_decode(out[:, inputs["input_ids"].shape[1]:], skip_special_toke
 ```
 
 ### Reasoning effort
-
 The chat template exposes three reasoning levels — `high` (default), `medium`, `low` — plus a thinking-off switch:
-
 ```python
 ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt",
                               reasoning_effort="medium")   # or enable_thinking=False
 ```
 
 ### Tool calling
-
 The chat template renders tool definitions for you. The model emits calls as `<tool_call><function=…><parameter=…>`, and you feed results back as a `tool` role message:
-
 ```python
 tools = [{
     "type": "function",
@@ -157,43 +159,43 @@ tools = [{
         },
     },
 }]
-
 msgs = [{"role": "user", "content": "What's the weather in Beijing right now?"}]
 ids = tok.apply_chat_template(msgs, tools=tools, add_generation_prompt=True,
                               return_tensors="pt").to(model.device)
 out = model.generate(ids, max_new_tokens=256)
 reply = tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True)
+
 # <tool_call>
+
 # <function=get_weather>
+
 # <parameter=city>
+
 # Beijing
+
 # </parameter>
+
 # </function>
+
 # </tool_call>
 
 # run the tool, append the result, generate the final answer
 msgs += [{"role": "assistant", "content": reply},
          {"role": "tool", "content": "Clear, 26°C, light northeasterly wind"}]
 ```
-
 Over the OpenAI API pass `tools=` the same way. The server returns the text above verbatim by default; to get structured `tool_calls`, configure sglang with a tool-call parser matching this format (likewise a reasoning parser, if you want the thinking span in `reasoning_content`).
 
 ### SGLang
-
 `serve.sh` starts a server from a stock public image, overlaying three files onto the image's `sglang` package and nothing else. See `sglang_patch/README.md`.
-
 ```bash
 docker run --gpus all --shm-size 64g -p 30001:8080 \
     -v /path/to/agnes-3.0-flash:/model \
     lmsysorg/sglang:nightly-dev-20260908-20ca564b \
     bash /agnes-3.0-flash/serve.sh --served-model-name Agnes-3.0-Flash
 ```
-
 `serve.sh` forwards extra command-line arguments to sglang, which is how `--served-model-name` takes effect; `--tp 2` works the same way. The server listens on port 8080 inside the container:
-
 ```python
 from openai import OpenAI
-
 client = OpenAI(api_key="EMPTY", base_url="http://localhost:30001/v1")
 response = client.chat.completions.create(
     model="Agnes-3.0-Flash",
@@ -203,11 +205,9 @@ response = client.chat.completions.create(
 )
 print(response.choices[0].message.content)
 ```
-
 Pass `stream=True` for streaming; `tools=` and `reasoning_effort=` are accepted the same way.
 
 ## Hardware Requirements
-
 | Resource | Recommendation |
 |---|---|
 | GPUs | 1 × NVIDIA H200 141 GB or NVIDIA H100 80 GB (or equivalent) at bf16 |
@@ -218,7 +218,6 @@ Pass `stream=True` for streaming; `tools=` and `reasoning_effort=` are accepted
 Actual context length and concurrency depend on KV-cache allocation, runtime overhead, and tensor-parallel configuration; validate the target workload on the intended hardware.
 
 ## Recommended Inference Settings
-
 | Setting | Recommended |
 |---|---|
 | `temperature` | 1.0 |
@@ -230,7 +229,6 @@ Actual context length and concurrency depend on KV-cache allocation, runtime ove
 These are the checkpoint's own `generation_config.json` defaults.
 
 ## Model Capabilities
-
 | Capability | Support |
 |---|---|
 | Advanced reasoning | Yes, with `high` / `medium` / `low` effort levels |
@@ -243,18 +241,16 @@ These are the checkpoint's own `generation_config.json` defaults.
 | OpenAI-compatible APIs | Chat Completions via sglang |
 
 ## License
-
 Released under the [Apache License 2.0](LICENSE).
 
 ## Citation
-
 ```bibtex
 @misc{agnes30flash2026,
-  title        = {Agnes-3.0-Flash},
+  title        = {Agnes-3.0-Flash Preview},
   author       = {{Agnes AI}},
   year         = {2026},
   month        = sep,
-  howpublished = {Open-weights model},
+  howpublished = {Open-weights preview checkpoint},
   url          = {https://agnes-ai.com/}
 }
-```
+```