Update README.md
Browse files
README.md
CHANGED
|
@@ -2,6 +2,8 @@
|
|
| 2 |
base_model: sarvamai/sarvam-1
|
| 3 |
library_name: peft
|
| 4 |
pipeline_tag: text-generation
|
|
|
|
|
|
|
| 5 |
tags:
|
| 6 |
- lora
|
| 7 |
- qlora
|
|
@@ -49,7 +51,7 @@ FinBot acts as a friendly, concise AI assistant for Indian stock market queries:
|
|
| 49 |
|---|---|
|
| 50 |
| **Base Model** | `sarvamai/sarvam-1` |
|
| 51 |
| **Method** | QLoRA (4-bit quantization via `bitsandbytes`) |
|
| 52 |
-
| **Dataset** | `shravankotagi/hinglish-finance-sft-v5-fixed` |
|
| 53 |
| **Train / Val Split** | 4,696 train / 825 validation examples |
|
| 54 |
| **Hardware** | Kaggle GPU |
|
| 55 |
| **LoRA Rank / Alpha** | `r=16` / `alpha=32` (`dropout=0.05`) |
|
|
@@ -59,6 +61,12 @@ FinBot acts as a friendly, concise AI assistant for Indian stock market queries:
|
|
| 59 |
| **Epochs / Total Steps**| 2 Epochs / 1,174 Steps |
|
| 60 |
| **Optimizer / LR** | `paged_adamw_8bit` / `1e-4` (Cosine Scheduler) |
|
| 61 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
---
|
| 63 |
|
| 64 |
## Training Loss Curve
|
|
@@ -133,4 +141,19 @@ with torch.no_grad():
|
|
| 133 |
)
|
| 134 |
|
| 135 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 136 |
-
print(response.split("[/INST]")[-1].strip())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
base_model: sarvamai/sarvam-1
|
| 3 |
library_name: peft
|
| 4 |
pipeline_tag: text-generation
|
| 5 |
+
datasets:
|
| 6 |
+
- shravankotagi/hinglish-finance-sft-v5-fixed
|
| 7 |
tags:
|
| 8 |
- lora
|
| 9 |
- qlora
|
|
|
|
| 51 |
|---|---|
|
| 52 |
| **Base Model** | `sarvamai/sarvam-1` |
|
| 53 |
| **Method** | QLoRA (4-bit quantization via `bitsandbytes`) |
|
| 54 |
+
| **Dataset** | [`shravankotagi/hinglish-finance-sft-v5-fixed`](https://huggingface.co/datasets/shravankotagi/hinglish-finance-sft-v5-fixed) |
|
| 55 |
| **Train / Val Split** | 4,696 train / 825 validation examples |
|
| 56 |
| **Hardware** | Kaggle GPU |
|
| 57 |
| **LoRA Rank / Alpha** | `r=16` / `alpha=32` (`dropout=0.05`) |
|
|
|
|
| 61 |
| **Epochs / Total Steps**| 2 Epochs / 1,174 Steps |
|
| 62 |
| **Optimizer / LR** | `paged_adamw_8bit` / `1e-4` (Cosine Scheduler) |
|
| 63 |
|
| 64 |
+
### Dataset Overview
|
| 65 |
+
The model was fine-tuned on [`shravankotagi/hinglish-finance-sft-v5-fixed`](https://huggingface.co/datasets/shravankotagi/hinglish-finance-sft-v5-fixed), a instruction dataset consisting of 5,521 instruction-response pairs covering:
|
| 66 |
+
- **Financial QA & Literacy:** Mutual funds, ETFs, portfolio rebalancing, and risk metrics.
|
| 67 |
+
- **Indian Taxation:** Capital gains (LTCG/STCG), ITR forms, and tax loss harvesting.
|
| 68 |
+
- **Market Sentiment & News:** Analysis of market trends, corporate actions, and regulatory updates in conversational Hinglish.
|
| 69 |
+
|
| 70 |
---
|
| 71 |
|
| 72 |
## Training Loss Curve
|
|
|
|
| 141 |
)
|
| 142 |
|
| 143 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 144 |
+
print(response.split("[/INST]")[-1].strip())
|
| 145 |
+
|
| 146 |
+
---
|
| 147 |
+
|
| 148 |
+
## Limitations & Disclaimers
|
| 149 |
+
|
| 150 |
+
- **Educational Purpose Only:** FinBot is an experimental model and does not provide certified financial, legal, or tax advice.
|
| 151 |
+
- **Dynamic Data:** Stock prices, tax slabs (LTCG/STCG), and exchange fees change frequently. Always verify live rates via official SEBI/NSE/BSE sources or consult a SEBI-registered financial advisor.
|
| 152 |
+
- **Scope:** Optimized specifically for Indian retail market contexts and Hinglish terminology; may underperform on non-Indian or purely technical corporate finance queries.
|
| 153 |
+
|
| 154 |
+
---
|
| 155 |
+
|
| 156 |
+
## Contact
|
| 157 |
+
|
| 158 |
+
- **Developer:** Shravan Kotagi
|
| 159 |
+
- **Hugging Face:** [@shravankotagi](https://huggingface.co/shravankotagi)
|