# Clawd Fable 8B Open Weights Released for Solana Agent Coding

> The 8-billion parameter model fine-tuned from Hermes-3-Llama-3.1 supports specialized tasks in Solana perps research and agent development while remaining distinct from live execution systems.

*Published 2026-09-18 · By Marcus Vance*

Clawd Fable 8B is a full merged Transformers model based on NousResearch/Hermes-3-Llama-3.1-8B with a LoRA adapter trained on combined Fable-5 code and trace datasets for Solana agent coding and reasoning.

The release provides Solana developers with a research-oriented model that integrates established base architecture with domain-specific adaptations. This approach allows targeted improvements in handling blockchain-specific instructions without requiring full retraining from scratch. The distribution includes both complete parameter sets and quantized formats to accommodate varied hardware environments.

## What background led to the Clawd Fable 8B release?

Solana projects have increasingly explored agent systems that interact directly with on-chain data for trading and strategy evaluation. Prior open models often lacked sufficient exposure to Solana transaction patterns and perpetuals mechanics. The current effort builds on existing fine-tuning techniques to close that gap while preserving general capabilities from the base model.

Community labs focused on Solana have published datasets that capture code generation and trace information relevant to decentralized finance workflows. These resources enable adapters that specialize model behavior for tasks such as generating agent scripts or evaluating risk parameters. The combination of public datasets with established base models creates a reproducible path for further iteration.

## What technical details define the model architecture and training?

The architecture begins with the Hermes-3-Llama-3.1-8B checkpoint and applies a LoRA adapter named solanaclawd/solana-nvidia-trading-factory-8b-lora. Training incorporated two primary datasets that supply both code examples and execution traces. The resulting merged model supports inference in standard Transformers format as well as GGUF quantization for broader accessibility.

Core components and specifications of the Clawd Fable 8B releaseComponentDetailsBase ModelNousResearch/Hermes-3-Llama-3.1-8BLoRA Adaptersolanaclawd/solana-nvidia-trading-factory-8b-loraTraining Datasetsarmand0e/claude-fable-5-claude-code combined with Glint-Research/Fable-5-tracesRelease FormatsFull merged Transformers weights and GGUFPrimary FocusSolana agent coding, perps research, and risk controls

## What performance metrics are associated with the release?

Reported metrics focus on token-level accuracy within the Solana trading factory domain. These figures provide a quantitative reference point for developers evaluating the model against specific workloads. Additional context on training dynamics such as loss curves appears in accompanying laboratory documentation.

Download activity on the primary repository reached 668 instances in the preceding month according to platform statistics. Such engagement indicates initial interest from the developer community working on Solana tooling.

## What are the intended applications for Clawd Fable 8B?

- Generating and debugging Solana agent code for on-chain interactions
- Conducting perps research and evaluating trading parameters
- Supporting cuFOLIO handoffs between strategy modules
- Designing paper trading strategies with integrated risk controls
- Facilitating developer experimentation with quantized inference on consumer hardware

The model remains positioned for research and development environments. Any deployment involving actual capital allocation requires separate execution infrastructure and compliance verification.

## What market and stakeholder implications follow from the release?

Open-weight releases of this type lower barriers for teams building custom agents on Solana. Stakeholders gain the ability to inspect weights and continue fine-tuning without licensing restrictions common in proprietary systems. The availability of both full and quantized versions broadens potential hardware targets from data-center GPUs to local workstations.

Ecosystem participants focused on perpetuals markets may integrate the model into simulation pipelines that test strategy variants before on-chain deployment. This separation between research inference and live execution aligns with existing risk management practices in decentralized finance.

## What expert reactions have accompanied the announcement?

The creator of the associated laboratory highlighted the transparency of the release process and the inclusion of training datasets. Such statements underscore a commitment to reproducible research within the Solana AI tooling community.

> I am proud to release: The official Solana clawd trading factory. Including the open source dataset. Everything we do is now onchain.8Bit🦞 (@0rdlibrary), Creator, 8 Bit Labs

## What developments are expected next for similar models?

Further iterations may incorporate additional Solana-specific datasets or refine the adapter for particular perpetuals instruments. Community contributions through the Hugging Face platform could produce derivative models that extend coverage to other on-chain protocols.

Continued monitoring of download trends and downstream usage will indicate whether the current specialization meets sustained demand. Integration with existing Solana development frameworks remains an area for potential tooling expansion.

## Sources

1. [The model is a full merged Transformers model based on Hermes-3-Llama-3.1-8B with LoRA adapter for Solana agent coding and reasoning.](https://huggingface.co/solanaclawd/solana-nvidia-trading-factory-8b)
2. [The Solana NVIDIA Trading Factory 8B LoRA reports train loss of 0.8064 and token accuracy of 85.5 percent.](https://8bitlabs.ai/)
3. [The creator announced the release of the official Solana clawd trading factory with open source dataset.](https://x.com/0rdlibrary/status/2068313538058150399)
4. [All models are now uploaded to Hugging Face... 5. **solanaclawd/clawd-fable** — Fable GGUF ... 4. **solanaclawd/solana-nvidia-trading-factory-8b-lora** — Trading Factory LoRA](https://x.com/0rdlibrary/status/2073498153953988835)

---
Source: https://aiintelreport.com/frontier-models/clawd-fable-8b-solana-model-release
Index: https://aiintelreport.com/llms.txt · Full text: https://aiintelreport.com/llms-full.txt