Upload folder using huggingface_hub

README.md

@@ -0,0 +1,59 @@
+---
+license: apache-2.0
+tags:
+- north-ml
+- flare-1
+- deltanet
+- hybrid-attention
+thumbnail: banner.png
+---
+
+# Fiber Lite — North ML Flare-1 (~1.6B)
+
+The mid Flare-1 tier: Qwen3.5-derived gated **DeltaNet + GQA hybrid**
+backbone (hidden 2048) with **AgentRegisters** — persistent learned state
+slots for agent-style memory. This is the full-precision training
+checkpoint (HF `transformers` format). The production 8-bit Kestrel build
+with the learned router, Fiber memory trunk, and search pipeline is at
+[North-ML1/Fiber](https://huggingface.co/North-ML1/Fiber).
+
+## Benchmarks (measured, same harness — Modal L4, greedy, Nov-2025)
+
+| Model | ARC-C | MMLU | GSM8K | HumanEval |
+|---|---|---|---|---|
+| ink (725M) | 63.8% | 32.0% | 30.0% | 37.5% |
+| **fiber-lite (1.6B)** | 72.9% | 38.0% | 16.0% | 37.5% |
+| fiber-pro (1.6B) | 80.9% | 40.0% | 14.0% | 48.4% |
+| Qwen3-0.6B (raw) | 49.2% | 29.0% | 4.0% | 4.7% |
+| Qwen3-1.7B (raw) | 72.9% | 39.5% | 43.0% | 68.8% |
+| SmolLM3-3B (raw) | 77.9% | 36.0% | 51.0% | 71.9% |
+
+Subsets: ARC-C n=199, MMLU n=200, GSM8K n=100, HumanEval n=64. Honest read:
+ties Qwen3-1.7B on ARC-C and MMLU at the same size class; trails on math
+and code — math reasoning is the known weak spot.
+
+## Specs
+
+| | |
+|---|---|
+| Parameters | ~1.63B (1.41B backbone + 158M Fiber trunk + registers + bridge) |
+| Layers | 24 |
+| Hidden | 2048 |
+| Attention | GQA 8 heads / 2 KV, DeltaNet:GQA = 3:1 |
+| Context | 262,144 tokens |
+| Vocab | 65,536 |
+
+## Interactive chat
+
+```bash
+pip install torch transformers safetensors huggingface_hub
+python3 chat.py --model North-ML1/fiber-lite
+```
+
+`chat.py` (in this repo) auto-downloads the weights, attaches the
+AgentRegisters, and starts a REPL with optional two-phase thinking.
+
+## Links
+
+- Architecture report: [Flare-1 paper](https://huggingface.co/spaces/North-ML1/flare-1-paper)
+- Runtime artifacts: [Ink-1](https://huggingface.co/North-ML1/Ink-1) · [Fiber](https://huggingface.co/North-ML1/Fiber) · [Fiber-pro](https://huggingface.co/North-ML1/Fiber-pro)

agent_register.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:794a2d3bbd503bb521b6dca8d4e3ff38f0b2ee935e348a6bac39ecf0c3ce8a34
+size 1188461

chat.py

@@ -0,0 +1,152 @@
+"""Interactive chat with a North ML Flare model (HF-format checkpoint).
+
+  python3 chat.py --model North-ML1/fiber-pro
+  python3 chat.py --model North-ML1/ink --no-think
+
+Requires: pip install torch transformers safetensors huggingface_hub
+Commands: /think on|off   /new   /quit
+"""
+import argparse, json, os, sys, threading, importlib.util
+
+IM_S = "<" + "|im_start" + "|>"
+IM_E = "<" + "|im_end" + "|>"
+TH_S = "<" + "think" + ">"
+TH_E = "<" + "/think" + ">"
+EOT = "<" + "|endoftext" + "|>"
+
+
+def pick_device():
+    import torch
+    if torch.cuda.is_available():
+        return "cuda", torch.bfloat16
+    if torch.backends.mps.is_available():
+        return "mps", torch.float16
+    return "cpu", torch.float32
+
+
+def load(repo):
+    import torch
+    from huggingface_hub import snapshot_download
+    from transformers import AutoModelForCausalLM, AutoTokenizer
+
+    local = snapshot_download(repo) if not os.path.isdir(repo) else repo
+    device, dtype = pick_device()
+    tok = AutoTokenizer.from_pretrained(local)
+    model = AutoModelForCausalLM.from_pretrained(local, dtype=dtype).to(device).eval()
+
+    arch_p = os.path.join(local, "starlight_arch.py")
+    reg_p = os.path.join(local, "agent_register.pt")
+    meta_p = os.path.join(local, "starlight.json")
+    if all(os.path.exists(p) for p in (arch_p, reg_p, meta_p)):
+        spec = importlib.util.spec_from_file_location("starlight_arch", arch_p)
+        arch = importlib.util.module_from_spec(spec)
+        spec.loader.exec_module(arch)
+        meta = json.load(open(meta_p))
+        regs = arch.attach_registers(model, arch.hidden_size_of(model),
+                                     slots=int(meta.get("slots", 16)))
+        regs.load_state_dict(torch.load(reg_p, map_location="cpu")["registers"])
+        regs.to(device=device, dtype=dtype)
+        print("registers attached (%s slots)" % meta.get("slots", 16))
+    print("loaded on %s (%s)" % (device, dtype))
+    return model, tok, device
+
+
+def render(history, think):
+    s = ""
+    for role, text in history:
+        s += IM_S + role + "\n" + text + IM_E + "\n"
+    s += IM_S + "assistant\n"
+    if think:
+        s += TH_S + "\n"
+    return s
+
+
+def gen_stream(model, tok, device, input_ids, cap, eos_ids):
+    """Generate with streaming stdout. Returns (text, full_ids)."""
+    import torch
+    from transformers import TextIteratorStreamer
+    streamer = TextIteratorStreamer(tok, skip_special_tokens=True, skip_prompt=True)
+    holder = {}
+
+    def _run():
+        holder["out"] = model.generate(
+            input_ids=input_ids, max_new_tokens=cap, do_sample=False,
+            repetition_penalty=1.05, eos_token_id=eos_ids,
+            pad_token_id=tok.eos_token_id, streamer=streamer)
+
+    t = threading.Thread(target=_run)
+    t.start()
+    text = ""
+    for chunk in streamer:
+        text += chunk
+        sys.stdout.write(chunk)
+        sys.stdout.flush()
+    t.join()
+    return text, holder["out"]
+
+
+def chat_turn(model, tok, device, history, think):
+    import torch
+    im_end = tok.convert_tokens_to_ids(IM_E)
+    eot = tok.convert_tokens_to_ids(EOT)
+    th_e = tok.convert_tokens_to_ids(TH_E)
+    eos = list({i for i in (tok.eos_token_id, im_end, eot) if i is not None})
+
+    prompt = render(history, think)
+    ids = tok(prompt, add_special_tokens=False, return_tensors="pt").input_ids.to(device)
+
+    if think:
+        sys.stdout.write("\n[think] ")
+        th_text, out1 = gen_stream(model, tok, device, ids, 220, eos + [th_e])
+        gen_ids = out1[0].tolist()
+        if th_e not in gen_ids[ids.shape[1]:]:
+            extra = tok(TH_E + "\n", add_special_tokens=False, return_tensors="pt")
+            ids2 = torch.cat([out1, extra.input_ids.to(device)], dim=1)
+        else:
+            ids2 = out1
+        sys.stdout.write("\n[answer] ")
+        ans_text, _ = gen_stream(model, tok, device, ids2, 500, eos)
+        sys.stdout.write("\n")
+        return ans_text
+    ans_text, _ = gen_stream(model, tok, device, ids, 500, eos)
+    sys.stdout.write("\n")
+    return ans_text
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--model", default="North-ML1/fiber-pro")
+    ap.add_argument("--no-think", action="store_true")
+    a = ap.parse_args()
+
+    print("downloading %s ..." % a.model)
+    model, tok, device = load(a.model)
+    think = not a.no_think
+    history = []
+    print("\nReady. Ask anything.   /think on|off   /new   /quit\n")
+    while True:
+        try:
+            q = input("you> ").strip()
+        except (EOFError, KeyboardInterrupt):
+            print()
+            break
+        if not q:
+            continue
+        if q in ("/quit", "/q", "exit"):
+            break
+        if q == "/new":
+            history = []
+            print("(cleared)\n")
+            continue
+        if q.startswith("/think"):
+            think = "off" not in q
+            print("think =", "on" if think else "off", "\n")
+            continue
+        history.append(("user", q))
+        ans = chat_turn(model, tok, device, history, think)
+        history.append(("assistant", ans.strip()))
+        sys.stdout.write("\n")
+
+
+if __name__ == "__main__":
+    main()

chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,75 @@
+{
+  "architectures": [
+    "Qwen3_5ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attn_output_gate": true,
+  "bos_token_id": null,
+  "dtype": "bfloat16",
+  "eos_token_id": 65503,
+  "full_attention_interval": 4,
+  "head_dim": 256,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5504,
+  "layer_types": [
+    "linear_attention",
+    "linear_attention",
+    "linear_attention",
+    "full_attention",
+    "linear_attention",
+    "linear_attention",
+    "linear_attention",
+    "full_attention",
+    "linear_attention",
+    "linear_attention",
+    "linear_attention",
+    "full_attention",
+    "linear_attention",
+    "linear_attention",
+    "linear_attention",
+    "full_attention",
+    "linear_attention",
+    "linear_attention",
+    "linear_attention",
+    "full_attention",
+    "linear_attention",
+    "linear_attention",
+    "linear_attention",
+    "full_attention"
+  ],
+  "linear_conv_kernel_dim": 4,
+  "linear_key_head_dim": 128,
+  "linear_num_key_heads": 16,
+  "linear_num_value_heads": 16,
+  "linear_value_head_dim": 128,
+  "mamba_ssm_dtype": "float32",
+  "max_position_embeddings": 262144,
+  "mlp_only_layers": [],
+  "model_type": "qwen3_5_text",
+  "mtp_num_hidden_layers": 1,
+  "mtp_use_dedicated_embeddings": false,
+  "num_attention_heads": 8,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
+  "pad_token_id": null,
+  "partial_rotary_factor": 0.25,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "mrope_interleaved": true,
+    "mrope_section": [
+      11,
+      11,
+      10
+    ],
+    "partial_rotary_factor": 0.25,
+    "rope_theta": 10000000,
+    "rope_type": "default"
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.18.0",
+  "use_cache": false,
+  "vocab_size": 65536
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": 65503,
+  "transformers_version": "5.18.0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da49bcd0acbe36d7d1776edd67c1fc43d16041fa13cf54351342dc03d9121005
+size 2826264920

starlight.json

@@ -0,0 +1,7 @@
+{
+  "product": "Starlight Pro 1",
+  "base": "Qwen/Qwen3.5-2B",
+  "step": 2000,
+  "arch": "agent-register",
+  "slots": 16
+}

starlight_arch.py

@@ -0,0 +1,70 @@
+"""Starlight agent memory.
+
+Qwen3.5 stays the backbone. After each full-attention layer, a causal register
+mixes a low-rank memory back into the residual stream. The output projection
+starts at zero, so step 0 matches Qwen. The input projection does not, so the
+register gets a gradient and can learn.
+"""
+from __future__ import annotations
+
+import torch
+from torch import nn
+
+
+class AgentRegister(nn.Module):
+    def __init__(self, hidden: int, slots: int = 16):
+        super().__init__()
+        self.slots = slots
+        self.down = nn.Linear(hidden, slots, bias=False)
+        self.up = nn.Linear(slots, hidden, bias=False)
+        self.gate = nn.Linear(hidden, slots)
+        nn.init.normal_(self.down.weight, std=0.02)
+        nn.init.zeros_(self.up.weight)
+        nn.init.zeros_(self.gate.weight)
+        nn.init.constant_(self.gate.bias, -2.0)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        gate = torch.sigmoid(self.gate(hidden_states))
+        memory = torch.cumsum(gate * self.down(hidden_states), dim=1)
+        return hidden_states + self.up(memory)
+
+
+def full_attention_indexes(model) -> list[int]:
+    config = model.config
+    kinds = list(getattr(config, "layer_types", None) or [])
+    if not kinds and getattr(config, "text_config", None) is not None:
+        kinds = list(getattr(config.text_config, "layer_types", None) or [])
+    return [i for i, kind in enumerate(kinds) if kind == "full_attention"]
+
+
+def attach_registers(model, hidden: int, slots: int = 16) -> nn.ModuleList:
+    indexes = full_attention_indexes(model)
+    if not indexes:
+        raise RuntimeError("Qwen config has no full_attention layers")
+    registers = nn.ModuleList(AgentRegister(hidden, slots) for _ in indexes)
+    device = next(model.parameters()).device
+    dtype = next(model.parameters()).dtype
+    registers.to(device=device, dtype=dtype)
+
+    def make_hook(register: AgentRegister):
+        def hook(_module, _inputs, output):
+            if isinstance(output, tuple):
+                return (register(output[0]),) + output[1:]
+            return register(output)
+
+        return hook
+
+    layers = model.model.layers
+    for index, register in zip(indexes, registers):
+        layers[index].register_forward_hook(make_hook(register))
+    return registers
+
+
+def hidden_size_of(model) -> int:
+    config = model.config
+    size = getattr(config, "hidden_size", None)
+    if not size and getattr(config, "text_config", None) is not None:
+        size = getattr(config.text_config, "hidden_size", None)
+    if not size:
+        raise RuntimeError("missing hidden size")
+    return int(size)

tokenizer_config.json

@@ -0,0 +1,33 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "fix_mistral_regex": true,
+  "image_token": "<|image_pad|>",
+  "is_local": true,
+  "local_files_only": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}