Add particle-1.6 weights (no SFT dataset)

README.md

@@ -13,15 +13,31 @@ tags:
 
 ~100M Llama-style chat model, trained from scratch (same architecture as [particle-1.0](https://huggingface.co/prathamkode/particle-1.0)).
 
-**Weights are not on this page yet.** Public weights stay at [prathamkode/particle-1.0](https://huggingface.co/prathamkode/particle-1.0) until the v1.6 SFT bar holds.
+Weights are MIT. The NCERT / teacher JSONL is **not** on this page.
+
+## Usage
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+repo = "prathamkode/particle-1.6"
+tok = AutoTokenizer.from_pretrained(repo)
+model = AutoModelForCausalLM.from_pretrained(repo)
+
+messages = [{"role": "user", "content": "hello"}]
+prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+ids = tok(prompt, return_tensors="pt")
+out = model.generate(**ids, max_new_tokens=64, do_sample=False)
+print(tok.decode(out[0], skip_special_tokens=False))
+```
 
 ## Status (2026-08-23)
 
 - Pretrain: FineWeb-Edu, ~2B tokens (same base as 1.0)
-- SFT v1.6 trained on AWS Spot (`g6e.12xlarge`, 4× L40S) from that pretrain checkpoint
-- Mix: ~37k NCERT Q&A + ~1k manners + fact seeds + 60k code-filtered [smol-smoltalk](https://huggingface.co/datasets/HuggingFaceTB/smol-smoltalk) (**94,681** pairs)
+- SFT from that pretrain on AWS Spot (`g6e.12xlarge`, 4× L40S)
+- Mix: ~37k NCERT Q&A + ~1k manners + fact seeds + 60k code-filtered [smol-smoltalk](https://huggingface.co/datasets/HuggingFaceTB/smol-smoltalk) (**94,681** pairs). Dataset files stay private.
 - Informal smoke: `hello` greets and stops; facts still miss (e.g. capital of India → Kolkata)
-- Full greedy bar not closed. SFT checkpoint stays off the Hub
+- Public v1 weights remain at [prathamkode/particle-1.0](https://huggingface.co/prathamkode/particle-1.0)
 
 ## Model details
 
@@ -29,7 +45,8 @@ tags:
 - **Parameters:** 109.5M (12 layers, 768 hidden, 12 heads)
 - **Context:** 2048 tokens
 - **Tokenizer:** custom byte-level BPE, 32k vocab
-- **License (when weights land):** MIT
+- **Precision:** Hub weights `bfloat16`
+- **License:** MIT
 
 ## Intended use
 
@@ -41,4 +58,4 @@ Very small capacity. Hallucinates. English-centric. No RLHF.
 
 ## Citation
 
-Cite FineWeb-Edu / smol-smoltalk if you use the data. Current public weights: [prathamkode/particle-1.0](https://huggingface.co/prathamkode/particle-1.0).
+Cite FineWeb-Edu / smol-smoltalk if you use the data.

chat_template.jinja

@@ -0,0 +1,5 @@
+{% for message in messages %}{% if message['role'] == 'user' %}<|user|>
+{{ message['content'] }}
+{% elif message['role'] == 'assistant' %}<|assistant|>
+{{ message['content'] }}{% endif %}{% endfor %}{% if add_generation_prompt %}<|assistant|>
+{% endif %}

config.json

@@ -0,0 +1,32 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": null,
+  "dtype": "bfloat16",
+  "eos_token_id": 0,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 2048,
+  "max_position_embeddings": 2048,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_key_value_heads": 12,
+  "pad_token_id": 1,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_parameters": {
+    "rope_theta": 10000.0,
+    "rope_type": "default"
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.15.0",
+  "use_cache": true,
+  "vocab_size": 32000
+}

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": 0,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 1,
+  "transformers_version": "5.15.0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:411dc07e553a0708d1d87a5e1a0cc580b08bb94bc812454f53864327bca83164
+size 219071992

tokenizer_config.json

@@ -0,0 +1,13 @@
+{
+  "backend": "tokenizers",
+  "bos_token": null,
+  "eos_token": "<|endoftext|>",
+  "extra_special_tokens": [
+    "<|user|>",
+    "<|assistant|>"
+  ],
+  "model_max_length": 2048,
+  "pad_token": "<|padding|>",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": "<|unk|>"
+}