Upload folder using huggingface_hub
Browse files- README.md +30 -13
- simbot.safetensors +3 -0
README.md
CHANGED
|
@@ -13,32 +13,47 @@ framework: pytorch
|
|
| 13 |
|
| 14 |
# SimBot GPT (Level 1)
|
| 15 |
|
| 16 |
-
|
|
|
|
| 17 |
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
-
##
|
| 25 |
-
|
|
|
|
| 26 |
- `tokenizer.json` — BPE tokenizer
|
| 27 |
-
- `
|
|
|
|
|
|
|
|
|
|
| 28 |
|
| 29 |
-
##
|
| 30 |
|
| 31 |
```python
|
| 32 |
-
import
|
|
|
|
| 33 |
from tokenizers import Tokenizer
|
| 34 |
from model.simbot import SIMGPT
|
| 35 |
-
import json
|
| 36 |
|
|
|
|
| 37 |
tokenizer = Tokenizer.from_file("tokenizer.json")
|
| 38 |
|
|
|
|
| 39 |
with open("config.json") as f:
|
| 40 |
cfg = json.load(f)
|
| 41 |
|
|
|
|
| 42 |
model = SIMGPT(
|
| 43 |
vocab_size=cfg["vocab_size"],
|
| 44 |
block_size=cfg["block_size"],
|
|
@@ -47,5 +62,7 @@ model = SIMGPT(
|
|
| 47 |
d_model=cfg["d_model"]
|
| 48 |
)
|
| 49 |
|
| 50 |
-
|
|
|
|
|
|
|
| 51 |
model.eval()
|
|
|
|
| 13 |
|
| 14 |
# SimBot GPT (Level 1)
|
| 15 |
|
| 16 |
+
SimBot GPT is a **from-scratch GPT-style language model** implemented in **PyTorch**.
|
| 17 |
+
This project is focused on **learning LLM internals**, not on instruction tuning or production use.
|
| 18 |
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
## Model Overview
|
| 22 |
+
|
| 23 |
+
- **Architecture:** Decoder-only Transformer (GPT-like)
|
| 24 |
+
- **Training Objective:** Causal Language Modeling
|
| 25 |
+
- **Dataset:** Domain-specific text (Simdega / regional data)
|
| 26 |
+
- **Purpose:** Educational (understanding how LLMs work internally)
|
| 27 |
+
|
| 28 |
+
⚠️ This is a **base language model**, not instruction-tuned and not grounded with RAG.
|
| 29 |
+
|
| 30 |
+
---
|
| 31 |
|
| 32 |
+
## Repository Contents
|
| 33 |
+
|
| 34 |
+
- `simbot.safetensors` — model weights (safe & HF-recommended format)
|
| 35 |
- `tokenizer.json` — BPE tokenizer
|
| 36 |
+
- `config.json` — model hyperparameters
|
| 37 |
+
- `model/simbot.py` — model architecture (PyTorch)
|
| 38 |
+
|
| 39 |
+
---
|
| 40 |
|
| 41 |
+
## Usage Example
|
| 42 |
|
| 43 |
```python
|
| 44 |
+
import json
|
| 45 |
+
from safetensors.torch import load_file
|
| 46 |
from tokenizers import Tokenizer
|
| 47 |
from model.simbot import SIMGPT
|
|
|
|
| 48 |
|
| 49 |
+
# Load tokenizer
|
| 50 |
tokenizer = Tokenizer.from_file("tokenizer.json")
|
| 51 |
|
| 52 |
+
# Load config
|
| 53 |
with open("config.json") as f:
|
| 54 |
cfg = json.load(f)
|
| 55 |
|
| 56 |
+
# Build model
|
| 57 |
model = SIMGPT(
|
| 58 |
vocab_size=cfg["vocab_size"],
|
| 59 |
block_size=cfg["block_size"],
|
|
|
|
| 62 |
d_model=cfg["d_model"]
|
| 63 |
)
|
| 64 |
|
| 65 |
+
# Load weights
|
| 66 |
+
state_dict = load_file("simbot.safetensors")
|
| 67 |
+
model.load_state_dict(state_dict)
|
| 68 |
model.eval()
|
simbot.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:025e9428d136cc01078962849d732c1ee63dd73899f995246893ca58e0ab4b97
|
| 3 |
+
size 48078808
|