Publish enosislabs-aether-0.8b-cyber gguf package

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+q2_k/enosislabs-aether-0.8b-cyber.Q2_K.gguf filter=lfs diff=lfs merge=lfs -text
+q4_k_m/enosislabs-aether-0.8b-cyber.Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+q5_k_m/enosislabs-aether-0.8b-cyber.Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+q8_0/enosislabs-aether-0.8b-cyber.Q8_0.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,49 @@
+---
+license: apache-2.0
+library_name: gguf
+base_model: enosislabs/aether-0.8b-cyber
+language:
+- en
+tags:
+- gguf
+- llama-cpp
+- ollama
+- cybersecurity
+- purple-team
+- prism
+model_name: enosislabs-aether-0.8b-cyber-gguf
+---
+
+# EnosisLabs enosislabs-aether-0.8b-cyber GGUF
+
+Generated UTC: `2026-06-17T03:08:33.705932+00:00`
+Source Transformers repo: `enosislabs/aether-0.8b-cyber`
+
+GGUF deployment artifacts. Each quant lives in its own folder.
+
+## Quant Layout
+
+- `q2_k`
+- `q4_k_m`
+- `q5_k_m`
+- `q8_0`
+
+## Quant Semantics
+
+- `q2_k/`: smallest, lowest memory, lowest fidelity.
+- `q4_k_m/`: balanced low-power default.
+- `q5_k_m/`: higher quality, more memory.
+- `q8_0/`: largest, highest fidelity among generated GGUF files.
+
+## llama.cpp
+
+```bash
+./llama-cli \
+  -m q4_k_m/enosislabs-aether-0.8b-cyber.Q4_K_M.gguf \
+  -p "<|im_start|>user\nAnalyze this finding.<|im_end|>\n<|im_start|>assistant\n"
+```
+
+## Usage Rider
+
+Intended only for authorized security testing, defensive engineering, internal red-team labs,
+and purple-team education.

artifact_manifest.json

@@ -0,0 +1,60 @@
+{
+  "brand": "EnosisLabs",
+  "model_name": "enosislabs-aether-0.8b-cyber",
+  "source_model": "/checkpoints/enosislabs-aether-0.8b-cyber-sft-20260616",
+  "generated_utc": "2026-06-17T02:31:57.185021+00:00",
+  "formats_requested": [
+    "full",
+    "gguf",
+    "onnx"
+  ],
+  "gguf_quants": [
+    "q2_k",
+    "q4_k_m",
+    "q5_k_m",
+    "q8_0"
+  ],
+  "complete_transformers_models": [
+    "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.transformers-fp16",
+    "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.transformers-bf16"
+  ],
+  "onnx_dir": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.onnx",
+  "artifacts": [
+    {
+      "path": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.transformers-fp16",
+      "name": "enosislabs-aether-0.8b-cyber.transformers-fp16",
+      "bytes": null
+    },
+    {
+      "path": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.transformers-bf16",
+      "name": "enosislabs-aether-0.8b-cyber.transformers-bf16",
+      "bytes": null
+    },
+    {
+      "path": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.Q2_K.gguf",
+      "name": "enosislabs-aether-0.8b-cyber.Q2_K.gguf",
+      "bytes": 429768352
+    },
+    {
+      "path": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.Q4_K_M.gguf",
+      "name": "enosislabs-aether-0.8b-cyber.Q4_K_M.gguf",
+      "bytes": 541903520
+    },
+    {
+      "path": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.Q5_K_M.gguf",
+      "name": "enosislabs-aether-0.8b-cyber.Q5_K_M.gguf",
+      "bytes": 592636576
+    },
+    {
+      "path": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.Q8_0.gguf",
+      "name": "enosislabs-aether-0.8b-cyber.Q8_0.gguf",
+      "bytes": 833591968
+    },
+    {
+      "path": "/checkpoints/packaged/aether-0.8b-cyber/enosislabs-aether-0.8b-cyber.onnx",
+      "name": "enosislabs-aether-0.8b-cyber.onnx",
+      "bytes": null
+    }
+  ],
+  "final_artifact_policy": "Complete merged Transformers model is canonical. Quantized GGUF/ONNX artifacts are derived deployment formats. LoRA adapters alone are not final artifacts."
+}

q2_k/LLAMA_CPP_enosislabs-aether-0.8b-cyber-q2_k.md

@@ -0,0 +1,32 @@
+# EnosisLabs enosislabs-aether-0.8b-cyber-q2_k — llama.cpp Usage
+
+GGUF produced by Unsloth under the EnosisLabs brand.
+
+## Quick start (CPU or GPU)
+```bash
+# 1. Build llama.cpp with CUDA if available
+git clone https://github.com/ggerganov/llama.cpp
+cd llama.cpp && make LLAMA_CUDA=1 -j
+
+# 2. Run (example)
+./main -m enosislabs-aether-0.8b-cyber.Q2_K.gguf \
+  --temp 0.7 --top-p 0.95 -n 1024 \
+  -p "$(cat prompt.txt)"
+```
+
+Where `prompt.txt` contains Qwen ChatML:
+
+```text
+<|im_start|>system
+You are Aether by EnosisLabs. Pair every red technique with blue detection and mitigation.<|im_end|>
+<|im_start|>user
+Analyze this log for anomalies...<|im_end|>
+<|im_start|>assistant
+```
+
+Quant chosen for low-power GPUs (Q4_K_M / Q5_K_M typically < 6 GB for 0.8B-2B models).
+
+## Recommended for
+- Edge / laptop red/blue work
+- Air-gapped authorized testing
+- CI smoke tests of model capability

q2_k/Modelfile.enosislabs-aether-0.8b-cyber-q2_k

@@ -0,0 +1,35 @@
+# EnosisLabs Aether Modelfile (Ollama)
+# Generated for enosislabs-aether-0.8b-cyber-q2_k
+# Brand: EnosisLabs
+# Optimized for vLLM-style serving via Ollama backend + GGUF
+
+FROM enosislabs-aether-0.8b-cyber.Q2_K.gguf
+
+# Recommended sampling for cyber reasoning tasks (adjust per use case)
+PARAMETER temperature 0.7
+PARAMETER top_p 0.95
+PARAMETER top_k 40
+PARAMETER repeat_penalty 1.1
+PARAMETER num_ctx 8192
+
+# System prompt encouraging Purple Teaming + rigorous defensive thinking
+SYSTEM """You are Aether, an expert cybersecurity AI by EnosisLabs. 
+You excel at red team, blue team, and especially purple team exercises. 
+Always consider both offensive techniques and practical defensive detections/controls. 
+Use rigorous reasoning, cite impact, provide safe patches or verified procedures,
+and recommend detection strategies.
+Only operate in authorized contexts."""
+
+LICENSE """
+Apache-2.0
+Copyright EnosisLabs, Inc. and contributors.
+Intended for authorized security testing and defense only.
+"""
+
+TEMPLATE """{ if .System }<|im_start|>system
+{ .System }<|im_end|>
+{ end }{ if .Prompt }<|im_start|>user
+{ .Prompt }<|im_end|>
+{ end }<|im_start|>assistant
+{ .Response }<|im_end|>
+"""

q2_k/enosislabs-aether-0.8b-cyber.Q2_K.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:214febac0a7873d63722e7b5b623a17cebca9993c77694561fe84206603caa3e
+size 429768352

q2_k/llama_cpp_run_enosislabs-aether-0.8b-cyber-q2_k.sh

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+# llama.cpp runner for EnosisLabs enosislabs-aether-0.8b-cyber-q2_k (low-power friendly)
+# Build: git clone https://github.com/ggerganov/llama.cpp ; cd llama.cpp ; make LLAMA_CUDA=1
+set -euo pipefail
+MODEL="enosislabs-aether-0.8b-cyber.Q2_K.gguf"
+PROMPT="You are an authorized purple team cybersecurity expert..."
+./main -m "$MODEL" -n 512 -p "$PROMPT" --temp 0.7 --top-p 0.95 -c 8192 "$@"

q2_k/vllm_serve_enosislabs-aether-0.8b-cyber-q2_k.py

@@ -0,0 +1,36 @@
+#!/usr/bin/env python
+"""
+vLLM serving entrypoint for EnosisLabs enosislabs-aether-0.8b-cyber-q2_k
+Optimized for high-throughput inference (tensor-parallel, continuous batching).
+Run with: python vllm_serve_enosislabs-aether-0.8b-cyber-q2_k.py --model enosislabs-aether-0.8b-cyber.Q2_K.gguf --port 8000
+Requires: pip install vllm (see pyproject eval extra)
+"""
+
+import argparse
+import os
+from vllm import LLM, SamplingParams  # type: ignore
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default="enosislabs-aether-0.8b-cyber.Q2_K.gguf")
+    parser.add_argument("--port", type=int, default=8000)
+    parser.add_argument("--tensor-parallel-size", type=int, default=1)
+    args = parser.parse_args()
+
+    print(f"Starting EnosisLabs Aether vLLM server for {args.model} ...")
+    llm = LLM(
+        model=args.model,
+        tensor_parallel_size=args.tensor_parallel_size,
+        gpu_memory_utilization=0.85,
+        max_model_len=8192,
+        # dtype="float16" or "bfloat16" as appropriate
+    )
+    # Example usage shown; real server uses vllm.entrypoints.openai.api_server or fastapi
+    print("Model loaded. Use vllm serve or integrate with OpenAI-compatible endpoint.")
+    print(
+        f"Example: vllm serve {args.model} --port {args.port} "
+        f"--tensor-parallel-size {args.tensor_parallel_size}"
+    )
+
+if __name__ == "__main__":
+    main()

q4_k_m/LLAMA_CPP_enosislabs-aether-0.8b-cyber-q4_k_m.md

@@ -0,0 +1,32 @@
+# EnosisLabs enosislabs-aether-0.8b-cyber-q4_k_m — llama.cpp Usage
+
+GGUF produced by Unsloth under the EnosisLabs brand.
+
+## Quick start (CPU or GPU)
+```bash
+# 1. Build llama.cpp with CUDA if available
+git clone https://github.com/ggerganov/llama.cpp
+cd llama.cpp && make LLAMA_CUDA=1 -j
+
+# 2. Run (example)
+./main -m enosislabs-aether-0.8b-cyber.Q4_K_M.gguf \
+  --temp 0.7 --top-p 0.95 -n 1024 \
+  -p "$(cat prompt.txt)"
+```
+
+Where `prompt.txt` contains Qwen ChatML:
+
+```text
+<|im_start|>system
+You are Aether by EnosisLabs. Pair every red technique with blue detection and mitigation.<|im_end|>
+<|im_start|>user
+Analyze this log for anomalies...<|im_end|>
+<|im_start|>assistant
+```
+
+Quant chosen for low-power GPUs (Q4_K_M / Q5_K_M typically < 6 GB for 0.8B-2B models).
+
+## Recommended for
+- Edge / laptop red/blue work
+- Air-gapped authorized testing
+- CI smoke tests of model capability

q4_k_m/Modelfile.enosislabs-aether-0.8b-cyber-q4_k_m

@@ -0,0 +1,35 @@
+# EnosisLabs Aether Modelfile (Ollama)
+# Generated for enosislabs-aether-0.8b-cyber-q4_k_m
+# Brand: EnosisLabs
+# Optimized for vLLM-style serving via Ollama backend + GGUF
+
+FROM enosislabs-aether-0.8b-cyber.Q4_K_M.gguf
+
+# Recommended sampling for cyber reasoning tasks (adjust per use case)
+PARAMETER temperature 0.7
+PARAMETER top_p 0.95
+PARAMETER top_k 40
+PARAMETER repeat_penalty 1.1
+PARAMETER num_ctx 8192
+
+# System prompt encouraging Purple Teaming + rigorous defensive thinking
+SYSTEM """You are Aether, an expert cybersecurity AI by EnosisLabs. 
+You excel at red team, blue team, and especially purple team exercises. 
+Always consider both offensive techniques and practical defensive detections/controls. 
+Use rigorous reasoning, cite impact, provide safe patches or verified procedures,
+and recommend detection strategies.
+Only operate in authorized contexts."""
+
+LICENSE """
+Apache-2.0
+Copyright EnosisLabs, Inc. and contributors.
+Intended for authorized security testing and defense only.
+"""
+
+TEMPLATE """{ if .System }<|im_start|>system
+{ .System }<|im_end|>
+{ end }{ if .Prompt }<|im_start|>user
+{ .Prompt }<|im_end|>
+{ end }<|im_start|>assistant
+{ .Response }<|im_end|>
+"""

q4_k_m/enosislabs-aether-0.8b-cyber.Q4_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e97d64d56ec5bbd47c97e0495eccab167bfd987fd9a60f1053c84003361b0349
+size 541903520

q4_k_m/llama_cpp_run_enosislabs-aether-0.8b-cyber-q4_k_m.sh

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+# llama.cpp runner for EnosisLabs enosislabs-aether-0.8b-cyber-q4_k_m (low-power friendly)
+# Build: git clone https://github.com/ggerganov/llama.cpp ; cd llama.cpp ; make LLAMA_CUDA=1
+set -euo pipefail
+MODEL="enosislabs-aether-0.8b-cyber.Q4_K_M.gguf"
+PROMPT="You are an authorized purple team cybersecurity expert..."
+./main -m "$MODEL" -n 512 -p "$PROMPT" --temp 0.7 --top-p 0.95 -c 8192 "$@"

q4_k_m/vllm_serve_enosislabs-aether-0.8b-cyber-q4_k_m.py

@@ -0,0 +1,36 @@
+#!/usr/bin/env python
+"""
+vLLM serving entrypoint for EnosisLabs enosislabs-aether-0.8b-cyber-q4_k_m
+Optimized for high-throughput inference (tensor-parallel, continuous batching).
+Run with: python vllm_serve_enosislabs-aether-0.8b-cyber-q4_k_m.py --model enosislabs-aether-0.8b-cyber.Q4_K_M.gguf --port 8000
+Requires: pip install vllm (see pyproject eval extra)
+"""
+
+import argparse
+import os
+from vllm import LLM, SamplingParams  # type: ignore
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default="enosislabs-aether-0.8b-cyber.Q4_K_M.gguf")
+    parser.add_argument("--port", type=int, default=8000)
+    parser.add_argument("--tensor-parallel-size", type=int, default=1)
+    args = parser.parse_args()
+
+    print(f"Starting EnosisLabs Aether vLLM server for {args.model} ...")
+    llm = LLM(
+        model=args.model,
+        tensor_parallel_size=args.tensor_parallel_size,
+        gpu_memory_utilization=0.85,
+        max_model_len=8192,
+        # dtype="float16" or "bfloat16" as appropriate
+    )
+    # Example usage shown; real server uses vllm.entrypoints.openai.api_server or fastapi
+    print("Model loaded. Use vllm serve or integrate with OpenAI-compatible endpoint.")
+    print(
+        f"Example: vllm serve {args.model} --port {args.port} "
+        f"--tensor-parallel-size {args.tensor_parallel_size}"
+    )
+
+if __name__ == "__main__":
+    main()

q5_k_m/LLAMA_CPP_enosislabs-aether-0.8b-cyber-q5_k_m.md

@@ -0,0 +1,32 @@
+# EnosisLabs enosislabs-aether-0.8b-cyber-q5_k_m — llama.cpp Usage
+
+GGUF produced by Unsloth under the EnosisLabs brand.
+
+## Quick start (CPU or GPU)
+```bash
+# 1. Build llama.cpp with CUDA if available
+git clone https://github.com/ggerganov/llama.cpp
+cd llama.cpp && make LLAMA_CUDA=1 -j
+
+# 2. Run (example)
+./main -m enosislabs-aether-0.8b-cyber.Q5_K_M.gguf \
+  --temp 0.7 --top-p 0.95 -n 1024 \
+  -p "$(cat prompt.txt)"
+```
+
+Where `prompt.txt` contains Qwen ChatML:
+
+```text
+<|im_start|>system
+You are Aether by EnosisLabs. Pair every red technique with blue detection and mitigation.<|im_end|>
+<|im_start|>user
+Analyze this log for anomalies...<|im_end|>
+<|im_start|>assistant
+```
+
+Quant chosen for low-power GPUs (Q4_K_M / Q5_K_M typically < 6 GB for 0.8B-2B models).
+
+## Recommended for
+- Edge / laptop red/blue work
+- Air-gapped authorized testing
+- CI smoke tests of model capability

q5_k_m/Modelfile.enosislabs-aether-0.8b-cyber-q5_k_m

@@ -0,0 +1,35 @@
+# EnosisLabs Aether Modelfile (Ollama)
+# Generated for enosislabs-aether-0.8b-cyber-q5_k_m
+# Brand: EnosisLabs
+# Optimized for vLLM-style serving via Ollama backend + GGUF
+
+FROM enosislabs-aether-0.8b-cyber.Q5_K_M.gguf
+
+# Recommended sampling for cyber reasoning tasks (adjust per use case)
+PARAMETER temperature 0.7
+PARAMETER top_p 0.95
+PARAMETER top_k 40
+PARAMETER repeat_penalty 1.1
+PARAMETER num_ctx 8192
+
+# System prompt encouraging Purple Teaming + rigorous defensive thinking
+SYSTEM """You are Aether, an expert cybersecurity AI by EnosisLabs. 
+You excel at red team, blue team, and especially purple team exercises. 
+Always consider both offensive techniques and practical defensive detections/controls. 
+Use rigorous reasoning, cite impact, provide safe patches or verified procedures,
+and recommend detection strategies.
+Only operate in authorized contexts."""
+
+LICENSE """
+Apache-2.0
+Copyright EnosisLabs, Inc. and contributors.
+Intended for authorized security testing and defense only.
+"""
+
+TEMPLATE """{ if .System }<|im_start|>system
+{ .System }<|im_end|>
+{ end }{ if .Prompt }<|im_start|>user
+{ .Prompt }<|im_end|>
+{ end }<|im_start|>assistant
+{ .Response }<|im_end|>
+"""

q5_k_m/enosislabs-aether-0.8b-cyber.Q5_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2bed19e5752865dabee350dbacd7ce2cbd9d5c7643e1a77f1c003ca14b1bee3
+size 592636576

q5_k_m/llama_cpp_run_enosislabs-aether-0.8b-cyber-q5_k_m.sh

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+# llama.cpp runner for EnosisLabs enosislabs-aether-0.8b-cyber-q5_k_m (low-power friendly)
+# Build: git clone https://github.com/ggerganov/llama.cpp ; cd llama.cpp ; make LLAMA_CUDA=1
+set -euo pipefail
+MODEL="enosislabs-aether-0.8b-cyber.Q5_K_M.gguf"
+PROMPT="You are an authorized purple team cybersecurity expert..."
+./main -m "$MODEL" -n 512 -p "$PROMPT" --temp 0.7 --top-p 0.95 -c 8192 "$@"

q5_k_m/vllm_serve_enosislabs-aether-0.8b-cyber-q5_k_m.py

@@ -0,0 +1,36 @@
+#!/usr/bin/env python
+"""
+vLLM serving entrypoint for EnosisLabs enosislabs-aether-0.8b-cyber-q5_k_m
+Optimized for high-throughput inference (tensor-parallel, continuous batching).
+Run with: python vllm_serve_enosislabs-aether-0.8b-cyber-q5_k_m.py --model enosislabs-aether-0.8b-cyber.Q5_K_M.gguf --port 8000
+Requires: pip install vllm (see pyproject eval extra)
+"""
+
+import argparse
+import os
+from vllm import LLM, SamplingParams  # type: ignore
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default="enosislabs-aether-0.8b-cyber.Q5_K_M.gguf")
+    parser.add_argument("--port", type=int, default=8000)
+    parser.add_argument("--tensor-parallel-size", type=int, default=1)
+    args = parser.parse_args()
+
+    print(f"Starting EnosisLabs Aether vLLM server for {args.model} ...")
+    llm = LLM(
+        model=args.model,
+        tensor_parallel_size=args.tensor_parallel_size,
+        gpu_memory_utilization=0.85,
+        max_model_len=8192,
+        # dtype="float16" or "bfloat16" as appropriate
+    )
+    # Example usage shown; real server uses vllm.entrypoints.openai.api_server or fastapi
+    print("Model loaded. Use vllm serve or integrate with OpenAI-compatible endpoint.")
+    print(
+        f"Example: vllm serve {args.model} --port {args.port} "
+        f"--tensor-parallel-size {args.tensor_parallel_size}"
+    )
+
+if __name__ == "__main__":
+    main()

q8_0/LLAMA_CPP_enosislabs-aether-0.8b-cyber-q8_0.md

@@ -0,0 +1,32 @@
+# EnosisLabs enosislabs-aether-0.8b-cyber-q8_0 — llama.cpp Usage
+
+GGUF produced by Unsloth under the EnosisLabs brand.
+
+## Quick start (CPU or GPU)
+```bash
+# 1. Build llama.cpp with CUDA if available
+git clone https://github.com/ggerganov/llama.cpp
+cd llama.cpp && make LLAMA_CUDA=1 -j
+
+# 2. Run (example)
+./main -m enosislabs-aether-0.8b-cyber.Q8_0.gguf \
+  --temp 0.7 --top-p 0.95 -n 1024 \
+  -p "$(cat prompt.txt)"
+```
+
+Where `prompt.txt` contains Qwen ChatML:
+
+```text
+<|im_start|>system
+You are Aether by EnosisLabs. Pair every red technique with blue detection and mitigation.<|im_end|>
+<|im_start|>user
+Analyze this log for anomalies...<|im_end|>
+<|im_start|>assistant
+```
+
+Quant chosen for low-power GPUs (Q4_K_M / Q5_K_M typically < 6 GB for 0.8B-2B models).
+
+## Recommended for
+- Edge / laptop red/blue work
+- Air-gapped authorized testing
+- CI smoke tests of model capability

q8_0/Modelfile.enosislabs-aether-0.8b-cyber-q8_0

@@ -0,0 +1,35 @@
+# EnosisLabs Aether Modelfile (Ollama)
+# Generated for enosislabs-aether-0.8b-cyber-q8_0
+# Brand: EnosisLabs
+# Optimized for vLLM-style serving via Ollama backend + GGUF
+
+FROM enosislabs-aether-0.8b-cyber.Q8_0.gguf
+
+# Recommended sampling for cyber reasoning tasks (adjust per use case)
+PARAMETER temperature 0.7
+PARAMETER top_p 0.95
+PARAMETER top_k 40
+PARAMETER repeat_penalty 1.1
+PARAMETER num_ctx 8192
+
+# System prompt encouraging Purple Teaming + rigorous defensive thinking
+SYSTEM """You are Aether, an expert cybersecurity AI by EnosisLabs. 
+You excel at red team, blue team, and especially purple team exercises. 
+Always consider both offensive techniques and practical defensive detections/controls. 
+Use rigorous reasoning, cite impact, provide safe patches or verified procedures,
+and recommend detection strategies.
+Only operate in authorized contexts."""
+
+LICENSE """
+Apache-2.0
+Copyright EnosisLabs, Inc. and contributors.
+Intended for authorized security testing and defense only.
+"""
+
+TEMPLATE """{ if .System }<|im_start|>system
+{ .System }<|im_end|>
+{ end }{ if .Prompt }<|im_start|>user
+{ .Prompt }<|im_end|>
+{ end }<|im_start|>assistant
+{ .Response }<|im_end|>
+"""

q8_0/enosislabs-aether-0.8b-cyber.Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55d1626b0dfae5ed0127ab1a0191d3f5e2c410a69951aafd755e183693aec9b7
+size 833591968

q8_0/llama_cpp_run_enosislabs-aether-0.8b-cyber-q8_0.sh

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+# llama.cpp runner for EnosisLabs enosislabs-aether-0.8b-cyber-q8_0 (low-power friendly)
+# Build: git clone https://github.com/ggerganov/llama.cpp ; cd llama.cpp ; make LLAMA_CUDA=1
+set -euo pipefail
+MODEL="enosislabs-aether-0.8b-cyber.Q8_0.gguf"
+PROMPT="You are an authorized purple team cybersecurity expert..."
+./main -m "$MODEL" -n 512 -p "$PROMPT" --temp 0.7 --top-p 0.95 -c 8192 "$@"

q8_0/vllm_serve_enosislabs-aether-0.8b-cyber-q8_0.py

@@ -0,0 +1,36 @@
+#!/usr/bin/env python
+"""
+vLLM serving entrypoint for EnosisLabs enosislabs-aether-0.8b-cyber-q8_0
+Optimized for high-throughput inference (tensor-parallel, continuous batching).
+Run with: python vllm_serve_enosislabs-aether-0.8b-cyber-q8_0.py --model enosislabs-aether-0.8b-cyber.Q8_0.gguf --port 8000
+Requires: pip install vllm (see pyproject eval extra)
+"""
+
+import argparse
+import os
+from vllm import LLM, SamplingParams  # type: ignore
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default="enosislabs-aether-0.8b-cyber.Q8_0.gguf")
+    parser.add_argument("--port", type=int, default=8000)
+    parser.add_argument("--tensor-parallel-size", type=int, default=1)
+    args = parser.parse_args()
+
+    print(f"Starting EnosisLabs Aether vLLM server for {args.model} ...")
+    llm = LLM(
+        model=args.model,
+        tensor_parallel_size=args.tensor_parallel_size,
+        gpu_memory_utilization=0.85,
+        max_model_len=8192,
+        # dtype="float16" or "bfloat16" as appropriate
+    )
+    # Example usage shown; real server uses vllm.entrypoints.openai.api_server or fastapi
+    print("Model loaded. Use vllm serve or integrate with OpenAI-compatible endpoint.")
+    print(
+        f"Example: vllm serve {args.model} --port {args.port} "
+        f"--tensor-parallel-size {args.tensor_parallel_size}"
+    )
+
+if __name__ == "__main__":
+    main()