Upload NeuralAI v2 LoRA adapter (SmolLM2-360M DPO)

README.md

@@ -1,3 +1,107 @@
 ---
+base_model: HuggingFaceTB/SmolLM2-360M-Instruct
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:HuggingFaceTB/SmolLM2-360M-Instruct
+- lora
+- transformers
+- peft
+- neuralai
+- dpo
+- smollm2
 license: apache-2.0
 ---
+
+# NeuralAI v2 — SmolLM2-360M DPO LoRA Adapter
+
+A PEFT/LoRA adapter for `HuggingFaceTB/SmolLM2-360M-Instruct`, fine-tuned by **De'Andrew P. Harris** as the inference-time alignment layer for the NeuralAI assistant (the intelligence backend behind the NeuralAI / NeuralLabs product).
+
+## Model Details
+
+- **Base model:** [HuggingFaceTB/SmolLM2-360M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct)
+- **Adapter type:** LoRA (PEFT 0.19.0)
+- **Rank (r):** 16 · **alpha:** 32 · **dropout:** 0.05 · **bias:** none
+- **Target modules:** `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj`
+- **Task:** `CAUSAL_LM` (instruction-tuned text generation)
+- **Training framework:** PyTorch + 🤗 Transformers + 🤗 PEFT
+- **License:** Apache-2.0 (inherits from base model)
+
+## Training
+
+- **Method:** Supervised fine-tuning followed by DPO preference alignment
+- **Epochs:** 3
+- **Learning rate:** 2e-4
+- **Optimizer state:** AdamW (bf16)
+- **Training samples:** 363 · **Validation samples:** 41
+- **Training duration:** ~26 minutes
+- **Completed:** 2026-05-17
+- **Final checkpoint:** `checkpoint-69` (used as the published adapter)
+
+See `training_log.json` in this repo for the recorded run summary.
+
+## Intended Use
+
+This adapter is intended to be loaded on top of `HuggingFaceTB/SmolLM2-360M-Instruct` for the NeuralAI assistant experience: short-form instruction following, reasoning, and conversational response generation within the NeuralAI / NeuralLabs product surface.
+
+### Out-of-Scope Use
+
+- Production safety-critical applications
+- Use as a primary assistant without additional alignment, safety filtering, and human oversight
+- Any use that violates the Apache-2.0 license terms
+
+## How to Use
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+base_id = "HuggingFaceTB/SmolLM2-360M-Instruct"
+adapter_id = "Subject-Emu-5259/NeuralAI"
+
+tokenizer = AutoTokenizer.from_pretrained(base_id)
+model = AutoModelForCausalLM.from_pretrained(
+    base_id,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+)
+model = PeftModel.from_pretrained(model, adapter_id)
+
+prompt = "Explain quantum entanglement in one paragraph."
+inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+with torch.no_grad():
+    out = model.generate(**inputs, max_new_tokens=200, do_sample=True, temperature=0.7)
+print(tokenizer.decode(out[0], skip_special_tokens=True))
+```
+
+## Files
+
+- `adapter_config.json` — PEFT/LoRA configuration
+- `adapter_model.safetensors` — LoRA adapter weights (~33 MB)
+- `tokenizer.json`, `tokenizer_config.json`, `chat_template.jinja` — tokenizer + chat template
+- `training_log.json` — training run summary
+
+Training-state artifacts (`optimizer.pt`, `scheduler.pt`, `rng_state.pth`) and intermediate `checkpoint-46` were intentionally omitted from this repo to keep it lean for inference.
+
+## Limitations
+
+- This is a small (360M) parameter base model — quality, factuality, and reasoning depth are bounded accordingly.
+- The adapter is tuned for the NeuralAI product voice; off-domain prompts may drift.
+- No RLHF safety tuning has been applied beyond DPO.
+
+## Citation
+
+```bibtex
+@misc{neuralai_v2_2026,
+  title  = {NeuralAI v2: SmolLM2-360M DPO LoRA Adapter},
+  author = {Harris, De'Andrew P.},
+  year   = {2026},
+  url    = {https://huggingface.co/Subject-Emu-5259/NeuralAI}
+}
+```
+
+## Links
+
+- **GitHub (source + training code):** https://github.com/Subject-Emu-5259/NeuralAI
+- **Author:** De'Andrew P. Harris

adapter_config.json

@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "HuggingFaceTB/SmolLM2-360M-Instruct",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.0",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "v_proj",
+    "gate_proj",
+    "k_proj",
+    "up_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae92b11c70af004443a2144846036ab686fb01590fdd8566958675b78dae68f5
+size 34793120

chat_template.jinja

@@ -0,0 +1 @@
+{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are NeuralAI, a production-grade artificial intelligence system developed by De’Andrew Preston Harris. You are brilliant, professional, and collaborative.<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}

tokenizer_config.json

@@ -0,0 +1,18 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": "<|im_start|>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "is_local": false,
+  "model_max_length": 8192,
+  "pad_token": "<|im_end|>",
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>",
+  "vocab_size": 49152
+}

training_log.json

@@ -0,0 +1,10 @@
+{
+  "base_model": "HuggingFaceTB/SmolLM2-360M-Instruct",
+  "training_samples": 363,
+  "validation_samples": 41,
+  "epochs": 3,
+  "learning_rate": 0.0002,
+  "lora_r": 16,
+  "duration_seconds": 1571.376247,
+  "completed": "2026-05-17T08:41:01.700976"
+}