Upload folder using huggingface_hub

README.md

@@ -0,0 +1,146 @@
+---
+language:
+- en
+tags:
+- north-ml
+- wind
+- wind-edge
+- generative-ai
+- language-model
+- veyrian
+- research
+pipeline_tag: text-generation
+---
+Wind Edge 1.7
+The Final Traditional LLM. Not the Final AI.
+Wind Edge 1.7 is a new language model developed by North ML, representing the final planned traditional large language model in our research roadmap.
+Wind Edge 1.7 concludes a chapter of North ML's development history, spanning dozens of experimental prototypes, model families, and research initiatives.
+However, this does not mean North ML is abandoning language models, generative AI, or text generation.
+Instead, we are transitioning toward an entirely new research direction: Veyrian Execution Generators (VEGs).
+Our goal is to move beyond the limitations of conventional architectures while continuing to develop AI systems that understand, reason, and generate text.
+Model Overview
+Property
+Description
+Model
+Wind Edge 1.7
+Organization
+North ML
+Family
+Wind
+Model Category
+Generative Language Model
+Generation
+1.7
+Research Focus
+Efficient language generation
+Status
+New model
+Future Research
+Veyrian Execution Generators
+Planned Successor
+Wind Flame 2.0
+The End of the Traditional Wind LLM Era
+North ML has developed numerous experimental models, architectures, and prototypes throughout its research history.
+Our work has explored different approaches to model efficiency, language understanding, reasoning, and accessible AI.
+Wind Edge 1.7 represents the final planned traditional LLM in this journey.
+It is not the end of the Wind family.
+It is the beginning of a transition toward something more ambitious.
+The next generation of Wind models will still be generative AI systems. They will simply investigate a different architectural foundation.
+Introducing Veyrian Execution Generators (VEGs)
+A Veyrian Execution Generator, abbreviated VEG, is a proposed new class of generative artificial intelligence models being researched by North ML.
+VEGs are designed to understand and generate natural language, perform reasoning, solve problems, and carry out complex intelligent tasks.
+Unlike traditional Transformer-based language models, VEGs aim to use novel computational architectures and execution mechanisms to pursue greater intelligence with fewer resources.
+Core Research Goals
+1. Frontier-Level Intelligence
+VEGs aim to achieve extremely advanced reasoning, knowledge processing, and problem-solving capabilities, potentially competing with future frontier AI systems.
+2. Native Text Generation
+VEGs remain generative AI models.
+They are intended to generate text directly, hold conversations, explain concepts, write code, and perform language-based tasks without requiring conversion into a Transformer.
+3. Improved Parameter Efficiency
+Our objective is to achieve greater capabilities using fewer parameters and less computation than comparable conventional architectures.
+4. Novel Architecture
+VEGs will explore computational approaches that differ from standard Transformer-based models.
+This research may include new processing structures, execution mechanisms, and approaches to representing and generating information.
+5. Architectural Convertibility
+A key research objective is to investigate whether VEG models can be converted into conventional Transformer-based LLMs.
+Such conversion could enable deployment across existing language model infrastructure while retaining useful capabilities developed in a VEG.
+6. Advanced Reasoning
+VEGs are intended to prioritize reasoning quality, complex problem-solving, and adaptive computation rather than focusing exclusively on parameter count.
+These objectives remain experimental and do not represent established performance claims.
+Wind Flame 2.0: The VEG Generation
+Wind Flame 2.0 is North ML's planned VEG model and a major step in our next research direction.
+It is intended to introduce a different approach to generative intelligence while retaining core capabilities associated with modern language models.
+Planned Capabilities
+Capability
+Wind Flame 2.0 Goal
+Text generation
+Native
+Natural language understanding
+Yes
+Conversational AI
+Yes
+Code generation
+Targeted
+Advanced reasoning
+Primary research goal
+Novel architecture
+VEG
+Transformer conversion
+Planned research capability
+Parameter efficiency
+Primary optimization goal
+Frontier intelligence
+Long-term objective
+VEG-to-Transformer Conversion
+One of the most ambitious objectives for Wind Flame 2.0 is the possibility of converting a trained VEG into a conventional Transformer-based LLM.
+The proposed process would allow a model to operate in two architectural forms:
+- Native VEG: A generative AI model using the Veyrian architecture.
+- Converted Transformer: A conventional Transformer-based language model derived from the VEG.
+The objective is to preserve as much useful intelligence and functionality as possible across the conversion.
+Whether this can be achieved efficiently and without significant capability loss is an open research question.
+Wind Edge 1.7 vs. Wind Flame 2.0
+Category
+Wind Edge 1.7
+Wind Flame 2.0
+Generation
+Traditional LLM era
+VEG era
+Architecture
+Conventional language-model approach
+Novel Veyrian architecture
+Text generation
+Yes
+Yes
+Reasoning
+Language-model capabilities
+Advanced reasoning target
+Transformer conversion
+Not a defining objective
+Planned
+Research direction
+Traditional LLM development
+Next-generation generative AI
+Wind Flame 2.0 is not intended to eliminate language generation.
+It is intended to explore how language-generating intelligence can be built differently.
+Our Vision
+North ML believes that the future of artificial intelligence should not depend exclusively on increasing parameter counts, training budgets, and infrastructure requirements.
+Our research aims to investigate whether new architectural designs can provide substantial improvements in intelligence, efficiency, and accessibility.
+With Wind Edge 1.7, we conclude our current traditional LLM roadmap.
+With Veyrian, we intend to explore a new generation of generative AI systems.
+Our ambition is not simply to build larger models. It is to build smarter models.
+Research and Development Status
+Wind Edge 1.7 is part of North ML's ongoing Wind development efforts.
+Technical specifications, training methodology, benchmark results, supported configurations, and model availability will be documented as they are confirmed.
+Veyrian Execution Generators and Wind Flame 2.0 are planned research initiatives. Their proposed architecture and capabilities have not yet been independently demonstrated.
+This documentation distinguishes current model information from future research objectives.
+About North ML
+North ML is an independent artificial intelligence research project focused on developing efficient, capable, and accessible AI systems.
+Our work includes experimental architectures, language models, model optimization, and research into new approaches to generative intelligence.
+Explore our models and research:
+[North ML on Hugging Face](https://huggingface.co/North-ML1)
+A New Era Begins
+Wind Edge 1.7 is the last planned traditional LLM from North ML.
+Wind Flame 2.0 is planned as the beginning of our VEG generation.
+We are not leaving generative AI behind.
+We are exploring what it could become.
+North ML — Beyond Conventional Intelligence.

chat_template.jinja

@@ -0,0 +1,4 @@
+{%- for m in messages %}<|im_start|>{{ m['role'] }}
+{{ m['content'] }}<|im_end|>
+{%- endfor %}{%- if add_generation_prompt %}<|im_start|>assistant
+{% endif %}

config.json

@@ -0,0 +1,50 @@
+{
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "dtype": "bfloat16",
+  "eos_token_id": 2,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 2048,
+  "max_window_layers": 28,
+  "model_type": "qwen3",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 18,
+  "num_key_value_heads": 4,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 500000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.57.3",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 32000
+}

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 1,
+  "do_sample": true,
+  "eos_token_id": 2,
+  "pad_token_id": 2,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.57.3"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:260e6a741ecc24a96eb87b91caca5bd16f7d520870c4eff860daa4660596c897
+size 409814944

special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92eeea783c9b51a3d0cf9756aa86d248b57b1645a7285610f21dcad9aede5fff
+size 556697

tokenizer_config.json

@@ -0,0 +1,44 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "</s>",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizerFast",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}