d4rkninja/tanpo-deals
d4rkninja/tanpo-deals is a 1.2 billion parameter, instruction-tuned language model developed by d4rkninja, based on LiquidAI/LFM2.5-1.2B-Instruct, with a 32,768-token context length. This compact, domain-specialized model is fine-tuned for B2B deal-closing and channel/distributor management tasks. It excels at qualification, objection handling, proposal generation, and partner motions, making it suitable for local and inexpensive deployment.
Loading preview...
Tanpo Deals: A Compact B2B Specialist
d4rkninja/tanpo-deals is a 1.2 billion parameter, instruction-tuned model derived from LiquidAI/LFM2.5-1.2B-Instruct, featuring a 32,768-token context length. It is part of the Tanpo family of compact, domain-specialized models designed for local and inexpensive deployment, focusing specifically on B2B deal-closing and channel management workflows.
Key Capabilities
- Deal Qualification: Assists with discovery, go/no-go scorecards, and identifying red flags.
- Objection Handling: Provides strategies for reframing objections and crafting conditional next steps.
- Proposal Generation: Helps outline commercial proposals, ROI analyses, and mutual action plans.
- Channel Enablement: Supports distributor recruiting, channel enablement, and wholesale pricing architecture drafts.
Performance & Training
The model was fine-tuned using LoRA (PEFT) via Unsloth on the d4rkninja/tanpo-deals-sft dataset, comprising approximately 10,000 chat SFT examples. Internal automated domain evaluation (DarkLab harness) showed a +7.3 percentage point improvement over the base model, achieving 95.2% on a specialized rubric for deals tasks. This evaluation is directional and not an industry benchmark.
Best For
- Discovery and qualification gates
- Objection handling and negotiation packages
- Commercial proposal outlines and mutual action plans
- Distributor recruiting and channel enablement
Limitations
As a specialized model, its quality may decrease outside the defined deals workflow. Its 1.2B parameter scale limits world knowledge and long-horizon reasoning compared to larger general-purpose models. Outputs are drafts and require human review, especially for legal or financial implications.