selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

ByteDance-Seed logo

Cola2

ByteDance-Seed·Instruct·released 6 Nov 2026

Unknown licence8.8B32K ctx ≈6.8 GB
Your GPU
hf download ByteDance-Seed/Cola2
Setup ↓

Cola2 is a continuous-latent language model that co-adapts semantic compression and discrete text generation. It combines a dynamic-compression VAE, a block-causal Diffusion Transformer (DiT), and a span-causal token decoder with latent feedback.

Notify me when

How much VRAM does Cola2 need?

At Q4_K_M Cola2 needs about 6.8 GB with an 8K context (6.8 GB at Q4, 19 GB at full precision). Pick your GPU to see what fits and how fast it runs.

--vram
any
--gpus
--quant

Needs: ≈6.8 GB · Q4_K_M · 8K context — Pick a GPU to check the fit.

Change quant, context or KV cache to compare
Advanced: context 8K · KV cache 16-bit
Context
KV cache

Smallest GPU for each quant and context (KV cache 16-bit)

Quantisation4K8K16K32K
Q3_K_M
Q4_K_M
Q5_K_M
Q6_K
Q8_0
FP16

≤16 GB card24–32 GB48–96 GBmore

QuantisationQualityWeightsKV cacheTotal
Q2_KHeavy lossSmallest, noticeably worse qualityHeavy loss2.7 GB 1.1 GB*≈4.4 GB
Q3_K_MNoticeable lossSmall, some quality lossNoticeable loss4.0 GB 1.1 GB*≈5.8 GB
Q4_K_MrecommendedSmall lossRecommended default for local useSmall loss5.0 GB 1.1 GB*≈6.8 GB
Q5_K_MNear-losslessNear-lossless for most tasksNear-lossless5.8 GB 1.1 GB*≈7.7 GB
Q6_KNear-losslessPractically losslessNear-lossless6.7 GB 1.1 GB*≈8.6 GB
Q8_0LosslessLossless in practice, twice the size of Q4Lossless8.7 GB 1.1 GB*≈11 GB
FP16 / BF16LosslessFull precision, as releasedLossless16 GB 1.1 GB*≈19 GB

Estimates for 8K tokens of context: weights + KV cache + runtime overhead. Speed assumes memory-bound decoding at your GPU’s bandwidth. Pick a GPU above and the fit column checks every row against it.

How to pick a quant

Take the biggest quant that fits with room left for context. Q4_K_M is the usual sweet spot. Below Q4 quality drops fast, above Q6 you mostly pay memory for nothing. Big models (70B+, large MoE) tolerate low quants better than small ones.

  • LosslessF16, Q8_0
  • Near-losslessQ6_K, Q5_K_M
  • Small lossQ5_K_S, Q4_K_M, IQ4_NL, IQ4_XS, Q4_K_S
  • Noticeable lossQ4_0, Q3_K_L, Q3_K_M, IQ3_M
  • Heavy lossIQ3_XXS, Q3_K_S, IQ2_M, Q2_K

Typical quality loss, not measured per model. Tiers come from KL divergence against FP16 on Llama 3 8B: llama.cpp perplexity README: LLaMA 3 8B scoreboard (KLD and ΔPPL vs FP16, Wikitext-2); llama.cpp quantize tool: ΔPPL per quant type @ Llama-3-8B

How much context fits?

Model quant
KV cache

At 8K context, Q4_K_M needs ≈6.8 GB(5.0 GB weights + 1.1 GB KV cache + 0.8 GB overhead). Pick a GPU above to see how much context fits.

05101K2K4K8K16K32KGB

Weights + overheadKV cache

Can the RTX 4060 run Cola2? →

Speed vs quality per quant

Smaller quants run faster and fit in less memory, but lose quality.

Q2_KQ3_K_MQ4_K_MQ5_K_MQ6_KQ8_0F16

Fits in VRAMTight fitCPU offload✕ won't runQuality is typical KL divergence vs FP16, not measured for this model.

Real-world speed of Cola2

Tokens per second measured by people running it at home. A GPU shows a median once 3 reports agree on the same setup. Reports far from our estimate wait for a review. Measure yours with llama-bench →

No medians yet. Be one of the first 3 to report a setup.

All reports

Run Cola2 locally

These commands load the original weights from Hugging Face.

Chat with the pipeline APIdocs
python
from transformers import pipeline

pipe = pipeline("text-generation", model="ByteDance-Seed/Cola2", dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Explain self-hosting in one paragraph."}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])

Install Transformers

About Cola2

Cola2 is a continuous-latent language model that co-adapts semantic compression and discrete text generation. It combines a dynamic-compression VAE, a block-causal Diffusion Transformer (DiT), and a span-causal token decoder with latent feedback. GitHub · 中文说明 · Paper · Project page The accompanying paper is titled “Cola2: Co-Adaptive Hierarchical Language Modeling with Continuous Semantic Compression and Balanced Discrete Generation.” Cola2 dynamically groups text into semantic spans, represents each span with a compact continuous latent, generates latent blocks with a DiT, and realizes those latents as discrete tokens. The released checkpoint was trained on 346.03B source tokens and reaches a reported 60.10 average across eight evaluation tasks. These are reported paper results; individual runs may differ because latent generation is stochastic. This repository contains: • eight sharded safetensors weight files; • tokenizer and generation configuration files; • Hugging Face Transformers remote-code modules; • normalized JSONL data for eight evaluation tasks; and • SHA256SUMS for release-file verification. The inference implementation, command-line tools, examples, benchmark runner, and documentation are maintained in the Cola2 GitHub repository. Training code is not included in this release. Click the preview to watch the full video · Open the interactive project page

Full model card on Hugging Face →

Alternatives to Cola2

All alternatives →
ornith-ai logo

Ornith-1.0-9B

ornith-ai·Chat

≈6.8 GB

Aloha! 🌺 Today, we are releasing Ornith-1.0, a self-improving family of open-source models for agentic coding.

9B256K ctx MIT
822.7KJun 2026gguf
ornith-ai logo

Ornith-1.5-9B

ornith-ai·Chat

≈7.2 GB

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

9.7B256K ctx MIT
570.4KAug 2026ollama
Qwen logo

Qwen3-8B

Qwen·Chat

≈6.5 GB

Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.

8.2B40K ctx Apache 2.0
6.1MApr 2025ollama
ibm-granite logo

granite-4.2-8b

ibm-granite·Chat

≈7.0 GB

Granite is a family of open-source large language models developed by IBM, designed for enterprise and research use. Granite models are built to be versatile, safe, and efficient — covering a range of sizes and capabilities from compact edge-deployable models to large-scale reasoning systems.

8.8B128K ctx Apache 2.0
59.9KAug 2026ollama
ibm-granite logo

granite-4.1-8b

ibm-granite·Chat

≈7.0 GB

Model Summary: Granite-4.1-8B is a 8B parameter long-context instruct model finetuned from Granite-4.1-8B-Base using a combination of open source instruction datasets with permissive license and internally collected synthetic datasets.

8.8B128K ctx Apache 2.0
82.2KApr 2026ollama
IFM logo

K2-Horizon-7B

IFM·Chat

≈7.0 GB

K2-Horizon-7B is the medium dense member of the K2-Horizon family: a 7B-core decoder-only model with a 512K context window.

9B512K ctx Apache 2.0
28.7KSept 2026gguf

Cola2 FAQ

How much VRAM does Cola2 need?
About 6.8 GB at Q4_K_M with an 8K context. Q8_0 needs about 11 GB and full precision about 19 GB. Longer contexts need more memory for the KV cache.
Can Cola2 run on the RTX 4060?
Yes. It fits at Q4_K_M (about 6.8 GB), at an estimated 28 tokens/s. It fits comfortably on the Apple M4 (16 GB), Radeon RX 6700 XT, RTX 3060 12GB.
Can I use Cola2 commercially?
Unclear. It is released under Unknown licence. No licence stated: assume all rights reserved. Check the licence on the model card before using it in a product.
What is the context length of Cola2?
32,768 tokens (32K). The full window needs far more memory than an 8K context.
How do I run Cola2 locally?
It runs with Transformers. Each has copy-paste examples in its docs.

Data from Hugging Face, the model card and public leaderboards. Memory and speed figures are estimates; real numbers depend on the runtime, context length and drivers. Always check the licence yourself before commercial use.