---

base_model: Qwen/Qwen3.8-27B

library_name: gguf

pipeline_tag: image-text-to-text

license: other

language:

• en
• fr

tags:

• unsloth
• lora
• qlora
• finance
• image-text-to-text
• vision
• video
• gguf
• llama.cpp
• ollama

model-index:

• name: Qwen3.8-27B-finance — GGUF

results:

- task:

type: text-generation

name: Causal language modeling

dataset:

name: finance SFT holdout

type: finance_dataset_master.v2.jsonl

metrics:

- type: loss

value: 0.644263

name: eval_loss

---

Qwen3.8-27B-finance — GGUF

Instruction-tuned finance assistant (macro, markets, trading, allocation).

| | |

|---|---|

| Base model | Qwen/Qwen3.8-27B |

| Domain | finance |

| Method | LoRA / QLoRA (Unsloth) · rank 8 · α 16 |

| Quantization at train | bf16 LoRA |

| Context | 4096 tokens |

| Dataset | finance_dataset_master.v2.jsonl · train 31517 / eval 1659 |

| GPU | NVIDIA RTX PRO 6000 Blackwell Server Edition (95.0 GiB) |

| Wall time | 167.1 min |

| Modalities kept | vision, video |

This checkpoint specialises the base model on expert finance SFT cards: macro regimes, rates and credit, hedge-fund styles, and market structure. It is a research / briefing aid, not a regulated advisor.

What changed vs the reference

Reference = the published base checkpoint Qwen/Qwen3.8-27B, plus the first in-run loss (LoRA ≈ 0 at step 0).

| Metric | Reference (base / first log) | This fine-tune | Δ |

|---|---:|---:|---:|

| Train loss (first → last logged) | 2.9676 | 0.7322 | -75.3% |

| Train loss (best) | — | 0.5686 | — |

| Eval loss (holdout, first → last) | 1.6722 | 0.6443 | -61.5% |

The first logged train loss is the closest in-run proxy for the base model (LoRA starts near zero). Option F, when executed, adds an independent holdout comparison against the frozen merged base.

Training data

• File: finance_dataset_master.v2.jsonl
• Path used at train time: /content/drive/MyDrive/finetuning/finance_dataset_master.v2.jsonl
• Split: 0.05 holdout, seed 42
• Format: chat-templated SFT (messages / instruction+output / ### Instruction + ### Response)

Training procedure

| Hyperparameter | Value |

|---|---|

| Epochs | 1 |

| Learning rate | 5e-05 |

| Warmup ratio | 0.05 |

| Device batch | 4 |

| Grad accum | 2 |

| Effective batch | 8 |

| Optim | adamw_8bit |

| Packing | True |

| LoRA targets | ['q_proj', 'k_proj', 'v_proj', 'o_proj', 'out_proj', 'gate_proj', 'up_proj', 'down_proj'] |

Intended use

Market briefing, explaining factor and style risk, and structuring notes for an investment committee.

Out of scope: Not investment advice. Do not use it as the sole input for live trading or client recommendations.

Multimodal

Kept towers: vision, video. Vision/audio layers were frozen during text SFT (vision=False, audio=False). Load the merged Transformers folder (or GGUF + mmproj) to keep image / video / audio.

Sample (post-train smoke test)

Prompt: Summarize in 4 bullets how a rate hike typically hits equity multiples and credit spreads.

system
Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.
user
Summarize in 4 bullets how a rate hike typically hits equity multiples and credit spreads.
assistant
<think>
We need answer concise 4 bullets. Need explain rate hike impact on equity multiples and credit spreads. Need likely: discount rates up, DCF multiples down, especially long-duration growth; earnings may lag, valuation compression. Credit: tighter financial conditions, higher funding costs, lower leverage capacity, higher spreads, especially high-yield and long-duration credit; but cyclical improvements can offset if hike is soft-landing. Need 4 bullets. Final only.
</think>

- Higher policy rates raise discount rates, which usually compress equity multiples, especially for long-duration growth and high-valuation stocks.
- Earnings may initially support prices, but if higher rates slow growth, multiple compression often dominates.
- Credit spreads tend to widen as borrowing costs rise, leverage capacity falls, and financial conditions tighten.
- High-yield and long-duration credit are typically more affected than investment-grade short-duration bonds.

How to use

Typical llama.cpp call:

llama-mtmd-cli -m Qwen3.8-27B-Q4_K_M.gguf --mmproj mmproj-Qwen3.8-27B-F32.gguf

Limitations

• Domain shift: quality drops outside the SFT topics.
• Eval above is holdout loss (and optional targeted checks). It is not a public leaderboard.
• The base model license and acceptable-use policy still apply.

License

other — inherit and respect the license of Qwen/Qwen3.8-27B.