Built on Jackrong's Opus-distilled model that already thinks like Claude, then fine-tuned on 16K ERP migration samples to reason about legacy ERP data.

Architecture

Qwen3.5-9B (base intelligence, 262K context)
  → Opus CoT distillation (structured reasoning: analyze → decompose → solve)
    → ERP fine-tune (pattern recognition, screen identification, transforms)

Capabilities

Give it ANY file from ANY legacy ERP system and it will:

Analyze column patterns and data formats
Identify the target Acumatica screen from contextual relationships
Map every column to the correct Acumatica field
Generate Polars transformation code for dirty data
List prerequisites and migration mode requirements

Supported Legacy Systems

Sage 100/50/300/Intacct, QuickBooks Desktop/Online, SAP B1/R3, NetSuite, Oracle EBS, Dynamics GP/NAV/AX/365BC, Xero, MYOB, Epicor, Infor SyteLine, Odoo, ERPNext, Fishbowl, and generic exports.

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "Abhijith93/erp-migration-phase1-opus-distilled-qwen3.5-9b",
    torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(
    "Abhijith93/erp-migration-phase1-opus-distilled-qwen3.5-9b", trust_remote_code=True,
)

Serving

bash

vllm serve Abhijith93/erp-migration-opus-distilled-qwen3.5-9b \
    --port 8000 --reasoning-parser qwen3 \
    --enable-auto-tool-choice --tool-call-parser qwen3_coder