You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

πŸŒ€ CelesteImperia Financial Engine v1.1 (Spiral)

The Quant-Reasoning LoRA for Indian Equities

This is a specialized Low-Rank Adaptation (LoRA) for Llama-3-8B, fine-tuned on a proprietary dataset of quantitative reasoning chains for the Indian stock market (NSE/BSE).


⚠️ IMPORTANT UPDATE (April 2026): > A significantly upgraded version of this model is now available. Indian Finance Quant v2.1 - DeepSeek Reasoning Edition has been released, featuring a "Dual-Brain" architecture capable of synthesizing technical indicators, live market sentiment, and complex regulatory/legal constraints using native <think> reasoning chains. Please migrate to v2.1 for all "Indian Finance Quant" deployments.


πŸš€ Key Features

  • Chain-of-Thought Reasoning: This engine uses tags to break down RSI, Support/Resistance, and Momentum before giving a verdict.
  • NSE/BSE Optimized: Understands the nuances of Indian market sentiment and large-cap stock behavior (e.g., Reliance, HDFC, TCS).
  • Lightweight & Fast: Optimized via Unsloth for 4-bit inference on consumer GPUs.

πŸ“‚ Repository Structure

To support various algorithmic trading and scraping pipelines, this repository contains three distinct formats:

  • /adapter: The raw LoRA adapter weights (.safetensors).
  • /full_model: The fully merged and sharded model weights (4 parts).
  • /gguf: A compressed Q4_K_M version optimized for CPU-bound inference (Hugging Face Spaces Free Tier).

πŸ’» Training Infrastructure

The v1.1 weights were trained locally in a dedicated WSL2 environment utilizing a dual-GPU architecture:

  • Primary Compute: NVIDIA RTX 3090
  • Secondary Accelerator: NVIDIA RTX A4000

πŸ› οΈ Usage & Quick Start

Option A: GPU Inference (Unsloth)

from unsloth import FastLanguageModel model, tokenizer = FastLanguageModel.from_pretrained("CelesteImperia/Llama-3-8B-Quant-Reasoning-v1.1")

Option B: CPU/Edge Inference (GGUF)

from huggingface_hub import hf_hub_download from llama_cpp import Llama

model_path = hf_hub_download( repo_id="CelesteImperia/Llama-3-8B-Quant-Reasoning-v1.1", filename="gguf/llama-3-8b.Q4_K_M.gguf" ) llm = Llama(model_path=model_path, n_ctx=2048)

πŸ“œ Terms of Use & Licensing

  • Base License: Subject to the Meta Llama-3 Community License.
  • Financial Data License: Reasoning layers are proprietary to CelesteImperia (CC BY-NC 4.0).
  • Non-Commercial Restriction: You may not use this model for commercial trading services or paid signal groups without consent.
  • Authenticity Tracking: All outputs contain visible and invisible steganographic watermarks to identify usage.
  • Disclaimer: This is an AI research tool. It does not provide financial advice. CelesteImperia is not responsible for financial losses.
Downloads last month
-
GGUF
Model size
8B params
Architecture
llama
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for CelesteImperia/Llama-3-8B-Quant-Reasoning-v1.1

Adapter
(709)
this model