TLF-7B-LLM-01: Upload fine-tuned LoRA adapters (2026-03-29)

0214600_adapters.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1057b241d2eede9f3ef30dc43e56fe4d5126e587ce8640a673a5f2fd66f7c2f
+size 109600779

0214700_adapters.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eeab3c90affc7d938c5164d82e9fa6271886e1b7711faec7e72bcfb3db037542
+size 109600779

0214800_adapters.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff8c3ee97e4ef0a92db281d8a88a43a554426e29c064af0b34b55bfc11e1a2a7
+size 109600779

0214900_adapters.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef3e386f80a697ac9357fb72db20b49e38edc1a1a8369b70857e4331c46ef806
+size 109600779

0215000_adapters.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77f72bd8d0b83a42420d2810e8a0c0d9a276259cff5ea5eca7c4d95b8122903d
+size 109600779

README.md

@@ -1,3 +1,75 @@
----
-license: mit
----
+---
+license: apache-2.0
+base_model: sarvamai/sarvam-1
+library_name: peft
+tags:
+- indic-nlp
+- dictionary
+- sanskrit
+- marathi
+- hindi
+- sft
+- lora
+---
+
+# TLF-7B-LLM-01
+
+## Model Description
+This model is a fine-tuned version of [sarvamai/sarvam-1](https://huggingface.co/sarvamai/sarvam-1) specialized for **Bilingual Indic Lexicography**. 
+It has been trained to provide structured morphological breakdowns, definitions, and regional translations for Sanskrit and other Indian regional languages.
+
+The training data was ingested through the **TLF Mega-Pipeline**, integrating structured dictionary databases (MSSQL) with unstructured regional texts to improve grammar and stylistic intelligence.
+
+## Intended Use
+- **Dictionary Lookups**: Providing high-accuracy definitions and etymologies.
+- **Morphological Analysis**: Breaking down complex Sanskrit/Indic root words.
+- **Regional Translation**: Translating word concepts across Marathi, Hindi, and English.
+
+## Training Hyperparameters
+The following hyperparameters were used during training:
+- **Engine**: MLX
+- **Learning Rate**: 2e-05
+- **Batch Size**: 1
+- **Gradient Accumulation**: 64
+- **Optimizer**: adamw_torch
+- **LR Scheduler**: cosine
+- **LoRA R**: 32
+- **LoRA Alpha**: 16
+- **Max Sequence Length**: 1024
+
+## Prompt Template
+To achieve the intended structured output, use the following prompt format:
+
+```text
+<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n{query} [/INST]
+```
+
+## Inference Example
+### Using MLX (Apple Silicon)
+```python
+import mlx_lm
+model, tokenizer = mlx_lm.load("YourAccount/TLF-7B-LLM-01")
+
+prompt = "Provide a comprehensive morphological breakdown for: 'Abacus'"
+# Use Sarvam/Llama template logic here
+response = mlx_lm.generate(model, tokenizer, prompt=prompt)
+print(response)
+```
+
+### Using Transformers
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+base_model = AutoModelForCausalLM.from_pretrained("sarvamai/sarvam-1")
+model = PeftModel.from_pretrained(base_model, "YourAccount/TLF-7B-LLM-01")
+tokenizer = AutoTokenizer.from_pretrained("sarvamai/sarvam-1")
+
+inputs = tokenizer(prompt, return_tensors="pt")
+outputs = model.generate(**inputs, max_new_tokens=256)
+print(tokenizer.decode(outputs[0]))
+```
+
+## Citation & Credits
+- **TLF Framework**: Architected for Unified Indic LLM Fine-tuning.
+- **Data Source**: Custom Dictionary & Regional Text Corpus.

adapter_config.json

@@ -0,0 +1,41 @@
+{
+    "adapter_path": "/Users/vishalkhapre/Documents/Code/TLF-7B-LLM-01/code/dict/../../raw/models/TLF-7B-MLX-01_sarvamai_sarvam-1",
+    "batch_size": 1,
+    "config": "/Users/vishalkhapre/Documents/Code/TLF-7B-LLM-01/code/dict/../../raw/models/TLF-7B-MLX-01_sarvamai_sarvam-1/mlx_config.yaml",
+    "data": "/Users/vishalkhapre/Documents/Code/TLF-7B-LLM-01/code/dict/../../raw/models/TLF-7B-MLX-01_sarvamai_sarvam-1/data",
+    "fine_tune_type": "lora",
+    "grad_accumulation_steps": 64,
+    "grad_checkpoint": true,
+    "iters": 215030,
+    "learning_rate": 2e-05,
+    "lora_parameters": {
+        "dropout": 0.1,
+        "rank": 32,
+        "scale": 16
+    },
+    "lr_schedule": null,
+    "mask_prompt": false,
+    "max_grad_norm": 0.05,
+    "max_seq_length": 1024,
+    "model": "sarvamai/sarvam-1",
+    "num_layers": 16,
+    "optimizer": "adam",
+    "optimizer_config": {
+        "adam": {},
+        "adamw": {},
+        "muon": {},
+        "sgd": {},
+        "adafactor": {}
+    },
+    "project_name": null,
+    "report_to": null,
+    "resume_adapter_file": null,
+    "save_every": 100,
+    "seed": 0,
+    "steps_per_eval": 200,
+    "steps_per_report": 10,
+    "test": false,
+    "test_batches": 500,
+    "train": true,
+    "val_batches": 25
+}

adapters.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77f72bd8d0b83a42420d2810e8a0c0d9a276259cff5ea5eca7c4d95b8122903d
+size 109600779

mlx_config.yaml

@@ -0,0 +1,16 @@
+adapter_path: /Users/vishalkhapre/Documents/Code/TLF-7B-LLM-01/code/dict/../../raw/models/TLF-7B-MLX-01_sarvamai_sarvam-1
+batch_size: 1
+data: /Users/vishalkhapre/Documents/Code/TLF-7B-LLM-01/code/dict/../../raw/models/TLF-7B-MLX-01_sarvamai_sarvam-1/data
+grad_accumulation_steps: 64
+grad_checkpoint: true
+iters: 215030
+learning_rate: 2.0e-05
+lora_parameters:
+  dropout: 0.1
+  rank: 32
+  scale: 16
+max_grad_norm: 0.05
+max_seq_length: 1024
+model: sarvamai/sarvam-1
+save_every: 100
+train: true