Instructions to use KikoCis/FastContext-1.0-4B-SFT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use KikoCis/FastContext-1.0-4B-SFT with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="KikoCis/FastContext-1.0-4B-SFT")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("KikoCis/FastContext-1.0-4B-SFT")
model = AutoModelForCausalLM.from_pretrained("KikoCis/FastContext-1.0-4B-SFT", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use KikoCis/FastContext-1.0-4B-SFT with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "KikoCis/FastContext-1.0-4B-SFT"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KikoCis/FastContext-1.0-4B-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/KikoCis/FastContext-1.0-4B-SFT

SGLang

How to use KikoCis/FastContext-1.0-4B-SFT with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "KikoCis/FastContext-1.0-4B-SFT" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KikoCis/FastContext-1.0-4B-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "KikoCis/FastContext-1.0-4B-SFT" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KikoCis/FastContext-1.0-4B-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use KikoCis/FastContext-1.0-4B-SFT with Docker Model Runner:
```
docker model run hf.co/KikoCis/FastContext-1.0-4B-SFT
```

KikoCis commited on Jul 2

Commit

8d52e0c

verified ·

1 Parent(s): 04d9e0e

card v2: banner + spec-sheet header + preservation story

Browse files

Files changed (1) hide show

README.md +68 -11

README.md CHANGED Viewed

@@ -1,21 +1,78 @@
 ---
 license: mit
-tags: [preserved, repository-exploration, subagent, coder, agentic, qwen3, 256k, long-context]
-language: [en]
 pipeline_tag: text-generation
 ---
-# FastContext-1.0-4B-SFT (preserved original weights)
-**Preserved copy of Microsoft's FastContext-1.0-4B-SFT**, which Microsoft **deleted from both HuggingFace and GitHub** (verified: 404 on both) about two weeks after open-sourcing it under MIT. Re-uploaded here so the weights stay available. **Own your AI.**
-## What it is
-A **repository-exploration subagent** for coding agents: invoked on-demand by your main agent, it fires **parallel read-only tool calls (READ / GLOB / GREP)** across a repo and returns **only the file paths + line ranges you need** as focused context — offloading file discovery so your coding agent burns far fewer tokens. Microsoft's announcement reported **~60% fewer tokens** and **+5.5% SWE-bench** *(their figures; source now deleted)*.
-Architecture: plain **Qwen3 dense 4B** (`Qwen3ForCausalLM`, 36 layers, **256K context**, MIT).
-## Quantized GGUF
-Long-context-imatrix GGUFs (any llama.cpp backend): **KikoCis/FastContext-1.0-4B-longctx-imatrix-GGUF**.
-## Credit
-Original © Microsoft, MIT license. This is an unmodified preservation mirror (weights unchanged).

 ---
 license: mit
+library_name: transformers
+tags:
+- preserved
+- repository-exploration
+- subagent
+- coder
+- agentic
+- qwen3
+- 256k
+- long-context
+language:
+- en
 pipeline_tag: text-generation
 ---
+![banner](banner.png)
+<div style="border:2px solid currentColor; font-family:ui-monospace,'SF Mono','Cascadia Mono',Consolas,'Liberation Mono',monospace;">
+  <div style="border-bottom:1px solid currentColor; padding:6px 12px; font-size:11px; letter-spacing:3px; text-transform:uppercase; opacity:0.7; text-align:center;">PRESERVED ORIGINAL // REMOVED BY MICROSOFT FROM HF + GITHUB // MIT</div>
+  <div style="padding:14px; display:flex; flex-wrap:wrap; align-items:center; justify-content:center; gap:18px;">
+<pre style="margin:0; flex:0 0 auto; font-family:ui-monospace,'SF Mono','Cascadia Mono',Consolas,monospace; font-size:9px; line-height:1.15; letter-spacing:0;">
+   microsoft/FastContext ──▶ 404
+   github.com/microsoft/FastContext ──▶ 404
+              │
+              ▼
+   ┌──────────────────────────┐
+   │  weights preserved here  │
+   │  bf16 · 8.0 GB · intact  │
+   └──────────────────────────┘
+     you can't un-open-source
+</pre>
+    <div style="flex:0 1 auto; max-width:100%; text-align:center;">
+      <div style="font-size:23px; font-weight:800; letter-spacing:1px;">FASTCONTEXT-1.0-4B-SFT</div>
+      <div style="font-size:12.5px; letter-spacing:1px; opacity:0.8; margin-top:5px;"><span style="white-space:nowrap;">PRESERVED ORIGINAL WEIGHTS</span> · <span style="white-space:nowrap;">QWEN3 DENSE 4B</span> · <span style="white-space:nowrap;">256K CONTEXT</span> · <span style="white-space:nowrap;">BF16 · 8.0 GB</span></div>
+    </div>
+  </div>
+  <table style="display:table; table-layout:fixed; width:100%; margin:0; border-collapse:collapse; font-family:ui-monospace,'SF Mono',Consolas,monospace; font-size:12px;">
+    <tr>
+      <td style="border-top:1px solid currentColor; border-right:1px solid currentColor; padding:8px 12px;"><div style="font-size:10px; letter-spacing:1px; opacity:0.6;">WEIGHTS</div><div style="font-weight:700;">BF16 · UNMODIFIED</div></td>
+      <td style="border-top:1px solid currentColor; border-right:1px solid currentColor; padding:8px 12px;"><div style="font-size:10px; letter-spacing:1px; opacity:0.6;">ARCH</div><div style="font-weight:700;">QWEN3 DENSE · 36L</div></td>
+      <td style="border-top:1px solid currentColor; border-right:1px solid currentColor; padding:8px 12px;"><div style="font-size:10px; letter-spacing:1px; opacity:0.6;">CONTEXT</div><div style="font-weight:700;">256K NATIVE</div></td>
+      <td style="border-top:1px solid currentColor; padding:8px 12px;"><div style="font-size:10px; letter-spacing:1px; opacity:0.6;">LICENSE</div><div style="font-weight:700;">MIT</div></td>
+    </tr>
+  </table>
+</div>
+> Microsoft open-sourced FastContext under MIT, then **deleted it from both HuggingFace and GitHub** about two weeks later (verified: 404 on both, 2026-07-02). MIT means preservation is legal — so here it is, unmodified. **Own your AI: a model on your disk can't be sunset by a quarterly review.**
+## 🔍 What it is
+A **repository-exploration subagent** for coding agents. Invoked on demand by your main agent, it fires **parallel read-only tool calls** (`READ` / `GLOB` / `GREP`) across a repo and returns **only the file paths + line ranges that matter**, as compact context. Your frontier coding agent stops wasting its context window (and your bill) crawling the file tree.
+Microsoft's (now-deleted) announcement reported **~60% fewer tokens** from the main coding agent and **+5.5% on SWE-bench** — their figures; the source no longer exists to cite.
+**Architecture**: plain `Qwen3ForCausalLM` dense 4B — 36 layers, 256K native context. No exotic modules; loads with standard `transformers`.
+## 🚀 Quick start
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+m = AutoModelForCausalLM.from_pretrained("KikoCis/FastContext-1.0-4B-SFT", torch_dtype="bfloat16", device_map="auto")
+tok = AutoTokenizer.from_pretrained("KikoCis/FastContext-1.0-4B-SFT")
+```
+**Don't want 8 GB?** Grab the **GGUF quants** (1.96–2.5 GB, long-context imatrix, retrieval-validated 30/30 vs this bf16):
+👉 [KikoCis/FastContext-1.0-4B-longctx-imatrix-GGUF](https://huggingface.co/KikoCis/FastContext-1.0-4B-longctx-imatrix-GGUF)
+## ⚠️ Good to know
+- It's a **scout, not a solver** — it finds and returns evidence; pair it with a main coding agent that writes the actual fix.
+- Upstream docs, harness code and issues were deleted along with the repos; usage conventions here come from the announcement and community mirrors.
+- Weights are **byte-identical** to the (re-uploaded) original — no fine-tuning, no edits.
+## 📚 Credit & license
+Model, weights, training: **© Microsoft** (MIT). This is a preservation mirror sourced via the `ShaunGves/FastContext-1.0-4B-SFT` re-upload after `microsoft/FastContext-1.0-4B-SFT` was removed. Nothing modified. Quantized companion + validation: KikoCis.