Tanpo Marketing
A compact marketing / GTM specialist (~1.2B) for hooks, campaigns, email, ICP messaging, and launch copy — built for local and inexpensive deployment, not as a general-purpose chatbot.
Creator: d4rkninja
Collection: Tanpo — Domain Specialists
Original upstream: LiquidAI/LFM2.5-1.2B-Instruct (~1.17B parameters, 32,768-token context, designed for edge/on-device deployment).
Fine-tuning: Unsloth-compatible loading of that checkpoint via hub id unsloth/LFM2.5-1.2B-Instruct (LoRA / PEFT).
This repository hosts the merged Transformers weights (LoRA merged into the base).
Overview
Tanpo is a family of compact domain-specialized business models for local / edge / inexpensive deployment. Different specialists cover different workflows. Tanpo Marketing is one specialist in that family (not a frontier or general-purpose model).
Related artifacts:
- LoRA adapter: d4rkninja/tanpo-marketing-LoRA
- GGUF quants: d4rkninja/tanpo-marketing-GGUF — prefer
Q4_K_Mwhen available
Best For
- Ad hooks, headlines, and short-form creative angles
- Cold email / outbound sequences and landing-page heroes
- ICP messaging, positioning, and campaign briefs
- Launch announcements, social captions, objection handling, and pricing/packaging copy
Not Designed For
- General coding, math, or knowledge-work outside marketing/GTM
- Legal review of claims, trademarks, or regulated advertising compliance
- Replacing brand strategy or human creative direction
- Frontier-scale open-ended reasoning across unrelated domains
Why a Specialist Model?
Small general models spread capacity thin. Tanpo Marketing is fine-tuned for marketing workflow shapes (structure, specificity, CTAs, ICP language) so a ~1.2B on-device model can be useful where a larger general model is unnecessary.
Evaluation
Internal automated evaluations. Treat as directional, not industry benchmarks.
Headline — DarkLab 20-task domain harness
| Model | Rubric overall |
|---|---|
Base LFM2.5-1.2B-Instruct | 93.7% |
| tanpo-marketing | 97.6% |
| Delta | +3.9 percentage points |
Artifacts: evaluation/ — marketing_tasks.jsonl, score_rubric.md, evaluation/README.md.
Methodology: Automated domain rubric on ~20 held-out marketing tasks; same prompts/decoding for base vs fine-tune.
Also published — public capability suite (33 tasks, strict rubric v2)
Different suite (not used for the headline table): base 94.3% → fine-tune 96.1% (+1.8 pp).
Artifacts: benchmark/ — scorecard.json, SCORECARD.md, tasks.jsonl, score_rubric.md, results/, charts/.
Limitations: Automated rubrics can reward structure over real-world quality; sample sizes are small; results may not transfer outside each task distribution.
Example Prompts
- User: Write three distinct Facebook ad hooks for a B2B analytics tool aimed at mid-market ops leaders. Each hook ≤12 words; include a concrete benefit.
- User: Draft a cold email (subject + body) for a founder selling a scheduling API to clinic ops managers. One clear CTA.
- User: Rewrite this landing hero for clarity and ICP fit. Product: inventory forecasting for independent retailers. Audience: owners with 1–5 stores.
Inference
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "d4rkninja/tanpo-marketing"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo,
torch_dtype="auto",
device_map="auto"
)
messages = [
{"role": "system", "content": 'You are Tanpo Marketing, a practical GTM copy assistant.'},
{"role": "user", "content": 'Write three distinct Facebook ad hooks for a B2B analytics tool aimed at mid-market ops leaders. Each hook ≤12 words; include a concrete benefit.'},
]
inputs = tokenizer.apply_chat_template(
messages, add_generation_prompt=True, return_tensors="pt"
).to(model.device)
out = model.generate(inputs, max_new_tokens=512)
print(tokenizer.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))
Training
Verified from published adapter configs / training artifacts (no unverified hyperparams):
| Field | Value |
|---|---|
| Method | LoRA (PEFT) via Unsloth FastLanguageModel on hub id unsloth/LFM2.5-1.2B-Instruct (Unsloth-compatible of LiquidAI/LFM2.5-1.2B-Instruct) |
LoRA rank (r) | 16 |
| LoRA alpha | 16 |
| LoRA dropout | 0 |
| Bias | none |
| Target modules | q_proj, k_proj, v_proj, out_proj, in_proj, w1, w2, w3 |
| Task type | CAUSAL_LM |
Merged via PEFT merge_and_unload into full weights in this repo.
Dataset
- d4rkninja/tanpo-marketing-sft — ~10k chat SFT examples (documented on the dataset card).
Limitations
- Specialized: quality drops outside the marketing workflow distribution.
- ~1.2B scale: limited world knowledge and long-horizon reasoning vs larger models.
- May refuse or redirect clearly off-domain asks. Creative taste is subjective; automated rubric gains do not equal human preference.
- Eval gains are rubric-based and directional only.
Responsible Use
Do not use for deceptive advertising, false claims, dark patterns, or undisclosed synthetic endorsements. Humans remain responsible for brand, legal, and platform-policy compliance.
License
license: other / license_name: lfm-1.0
Tanpo merged and GGUF weights are derivatives of LiquidAI/LFM2.5-1.2B-Instruct under the LFM Open License v1.0 (including the commercial Threshold of approximately $10M annual revenue). See the base model card and its LICENSE file. Credit: LiquidAI. Do not treat this stack as Apache-2.0.
Tanpo Family
Tanpo is a family of compact domain-specialized models for focused business workflows.
This specialist is available as:
- Merged:
d4rkninja/tanpo-marketing - LoRA:
d4rkninja/tanpo-marketing-LoRA - GGUF:
d4rkninja/tanpo-marketing-GGUF
Browse all Tanpo specialists: Tanpo — Domain Specialists