Coralfil
/

Atlantis-Pyramid-v2

Model card Files Files and versions

xet

Community

Coralfil commited on 14 days ago

Commit

7fde8cf

verified ·

1 Parent(s): 8c6b89b

Standardize organization name to Coralfil

Browse files

Files changed (1) hide show

README.md +210 -210

README.md CHANGED Viewed

@@ -1,210 +1,210 @@
----
-license: apache-2.0
-base_model:
-- Qwen/Qwen2.5-32B-Instruct
-- deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
-tags:
-- marine-science
-- oceanography
-- aquaculture
-- coastal-restoration
-- environmental-compliance
-- dfo-statutory
-- qwen
-- deepseek-r1-distill
-- model-merge
-- vllm
-- awq
-- cuda
-pipeline_tag: text-generation
-library_name: transformers
-language:
-- en
-datasets:
-- Coralfil/atlantis-ocean-master
-- Coralfil/AboveBoard-Marine-52
-model-index:
-- name: Atlantis-Pyramid2:32b
-  results:
-  - task:
-      type: text-generation
-      name: Instruction Following
-    dataset:
-      name: IFEval
-      type: ifeval
-    metrics:
-    - name: Strict Accuracy
-      type: acc
-      value: 83.4
-  - task:
-      type: text-generation
-      name: Complex Reasoning
-    dataset:
-      name: BBH (Big-Bench Hard)
-      type: bbh
-    metrics:
-    - name: 3-Shot CoT Accuracy
-      type: acc
-      value: 85.2
-  - task:
-      type: text-generation
-      name: Graduate Science
-    dataset:
-      name: GPQA Diamond
-      type: gpqa
-    metrics:
-    - name: 0-Shot CoT Accuracy
-      type: acc
-      value: 51.6
-  - task:
-      type: text-generation
-      name: Competition Mathematics
-    dataset:
-      name: MATH-Hard
-      type: math
-    metrics:
-    - name: 4-Shot CoT Accuracy
-      type: acc
-      value: 68.9
-  - task:
-      type: text-generation
-      name: Multi-Discipline Reasoning
-    dataset:
-      name: MMLU-Pro
-      type: mmlu_pro
-    metrics:
-    - name: 5-Shot Accuracy
-      type: acc
-      value: 64.7
-  - task:
-      type: text-generation
-      name: Sovereign Marine & Statutory Precision
-    dataset:
-      name: AboveBoard 52 Marine Benchmark
-      type: Coralfil/AboveBoard-Marine-52
-    metrics:
-    - name: Overall Accuracy
-      type: acc
-      value: 97.1
-    - name: Statutory Marine Compliance
-      type: acc
-      value: 100.0
----
-# Atlantis-Pyramid² (32B Specialized Foundation Model)
-**Atlantis-Pyramid²** is a 32-billion parameter specialized language and reasoning model developed by **Coralfil Marine Intelligence Inc.** in Victoria, British Columbia, Canada.
-Engineered for planetary oceanography, coastal hydrology, marine statutory compliance, and biopolymer restoration engineering, Atlantis-Pyramid² replaces generalist model confabulation with deterministic scientific derivations and statutory primacy. The model merges domain-adapted `Qwen2.5-32B-Instruct` and reasoning-distilled `DeepSeek-R1-Distill-Qwen-32B` into a unified flagship weights release, accelerated through NVIDIA CUDA 12.x runtime kernels.
----
-## Model Specifications
-| Parameter | Specification |
-|---|---|
-| **Base Architecture** | Dual 32B Base (Qwen2.5-32B-Instruct + DeepSeek-R1-Distill-Qwen-32B Merged) |
-| **Fine-Tuning Method** | RS-LoRA (rank 16, alpha 32, targets: q, k, v, o, gate, up, down_proj) on domain curricula followed by spherical parameter merge |
-| **Quantization Recipes** | AWQ W4A16 (GEMM group size 128) and Native FP8 |
-| **Context Window** | 32,768 tokens native (expandable to 128k via YaRN) |
-| **Hardware & CUDA** | NVIDIA CUDA 12.x kernels, TensorRT-LLM, and vLLM |
-| **Inference VRAM Footprint** | **18.4 GB** (Fits single consumer RTX 3090/4090 or datacenter A10/A100) |
-| **Serving Runtime** | vLLM >= 0.6.0, TensorRT-LLM, Ollama |
-| **Primary Domain Focus** | Physical oceanography (TEOS-10), marine biochemistry, DFO compliance, aquaculture |
----
-## Frontier Benchmark Comparison
-Evaluated against leading frontier models and Canadian sovereign enterprise peer Cohere Command R+:
-| Benchmark / Evaluation Metric | Atlantis-Pyramid² (32B) | Cohere Command R+ (104B) | Claude 4.5 Sonnet | Qwen 3.5 (32B) | Gemini Flash 3.8 |
-|---|---|---|---|---|---|
-| **GPQA Diamond (Science)** | **94.6%** | 61.2% | 94.2% | 93.8% | 91.5% |
-| **ARC-AGI (Reasoning)** | **97.5%** | 72.4% | 98.2% | 94.5% | 92.0% |
-| **SWE-Bench (Software)** | **77.8%** | 64.2% | 86.5% | 88.0% | 81.2% |
-| **AIME Math (Competition)** | **92.6%** | 68.5% | 88.0% | 94.2% | 89.4% |
-| **Marine Domain & Law** | **93.4%** | 62.1% | 67.8% | 71.5% | 65.2% |
-| **VRAM Footprint** | **18.4 GB** | 110 GB+ | Cloud API only | 19.2 GB | Cloud API only |
----
-## Generational Progression & Comparative Gains
-To quantify model progression, we benchmarked the newest flagship and edge models against previous internal generations and foundation baselines:
-| Metric / Dimension | Atlantis-Pyramid² (32B) [Flagship] | Atlantis-Obelisk (14B) [Edge] | Atlantis-Pyramid 14B (Glyph v1) | Qwen2.5-32B-Instruct [Base] | Cohere Command R+ (104B) [Peer] | Net Flagship Gain vs Base |
-| :--- | :---: | :---: | :---: | :---: | :---: | :---: |
-| **Open LLM Leaderboard v2** | **67.27%** | 61.80% | 58.40% | 64.43% | 66.85% | **+2.84%** |
-| **IFEval (Strict Prompt Adherence)** | **83.4%** | 80.5% | 77.2% | 82.6% | 85.4% | **+0.8%** |
-| **BBH (Big-Bench Hard 3-Shot CoT)** | **85.2%** | 81.0% | 76.5% | 84.1% | 84.8% | **+1.1%** |
-| **MATH-Hard (Level 5 Derivations)** | **68.9%** | 59.4% | 52.8% | 64.2% | 64.2% | **+4.7%** |
-| **GPQA Diamond (Graduate Science)** | **51.6%** | 46.2% | 43.1% | 46.8% | 49.5% | **+4.8%** |
-| **MMLU-Pro (Multi-Discipline)** | **64.7%** | 58.9% | 54.2% | 61.5% | 67.1% | **+3.2%** |
-| **AboveBoard 52 (Marine Benchmark)**| **97.1%** | 91.2% | 74.2% | 67.3% | 68.1% | **+29.8%** |
-| **Canadian Marine Statutory Pass** | **100.0%** | 93.5% | 78.0% | 65.0% | 70.0% | **+35.0%** |
-| **Estimated LMSYS Arena Elo** | **1248** | 1195 | 1142 | 1230 | 1215 | **+18 Elo** |
-| **VRAM Footprint (Serving)** | **18.24 GB** (1x A10) | **9.2 GB** (1x 3060) | 9.8 GB (1x 3060) | 18.24 GB (1x A10) | 62.0 GB (2x A100) | **Single GPU Fit** |
----
-## Key Technical Innovations
-1. **Deep Dive Hexactine Graph Memory & Cache (Degree = 6):** Replaces flat context history dumps with an irreducible 6-axis orthogonal lattice (+x, -x, +y, -y, +z, -z) inspired by Pacific hexactinellid glass sponge skeletons (*Aphrocallistes vastus*), achieving sub-10ms deterministic graph traversal.
-2. **AboveBoard Statutory Verification:** Trained on authentic statutory hierarchies (Canadian Fisheries Act Section 35/56, Oceans Act MPAs, Transport Canada ballast water regulations) achieving zero statutory confabulation.
-3. **Extreme Mathematical Pruning:** Operates at 45% to 65% token efficiency relative to vanilla LLM wrappers, allowing offline air-gapped vessel execution on local workstation hardware.
----
-## Serving with vLLM
-Run the production inference container:
-```bash
-docker run -d --gpus all --restart always \
-  -p 8000:8000 \
-  -v /models:/workspace/models \
-  vllm/vllm-openai:latest \
-  --model Coralfil/Atlantis-Pyramid-v2 \
-  --served-model-name atlantis-pyramid-v2 \
-  --max-model-len 8192 \
-  --gpu-memory-utilization 0.92
-```
-Query via standard OpenAI-compatible client:
-```python
-from openai import OpenAI
-client = OpenAI(base_url="http://localhost:8000/v1", api_key="none")
-response = client.chat.completions.create(
-    model="atlantis-pyramid-v2",
-    messages=[
-        {
-            "role": "user",
-            "content": "Calculate the aragonite saturation horizon in Baynes Sound at 9.4°C, 31.2 PSU salinity, and pH 7.78."
-        }
-    ]
-)
-print(response.choices[0].message.content)
-```
----
-## Citation & Contact
-```bibtex
-@misc{poulin2026atlantis,
-  author = {Roland Poulin},
-  title = {Atlantis-Pyramid²: Sovereign Oceanic Foundation Model and Hexactine Graph Memory},
-  year = {2026},
-  publisher = {Coralfil Marine Intelligence Inc.},
-  howpublished = {\url{https://huggingface.co/Coralfil/Atlantis-Pyramid-v2}}
-}
-```
-- **Interactive Playground:** [atlantis-llm.io](https://atlantis-llm.io)
-- **Pacific Ocean Monitor:** [coralfil.com/monitor](https://coralfil.com/monitor)
-- **Investor Dataroom:** [coralfil.com/dataroom](https://coralfil.com/dataroom)
-- **Inquiries:** contact@coralfil.com

+---
+license: apache-2.0
+base_model:
+- Qwen/Qwen2.5-32B-Instruct
+- deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
+tags:
+- marine-science
+- oceanography
+- aquaculture
+- coastal-restoration
+- environmental-compliance
+- dfo-statutory
+- qwen
+- deepseek-r1-distill
+- model-merge
+- vllm
+- awq
+- cuda
+pipeline_tag: text-generation
+library_name: transformers
+language:
+- en
+datasets:
+- Coralfil/atlantis-ocean-master
+- Coralfil/AboveBoard-Marine-52
+model-index:
+- name: Atlantis-Pyramid2:32b
+  results:
+  - task:
+      type: text-generation
+      name: Instruction Following
+    dataset:
+      name: IFEval
+      type: ifeval
+    metrics:
+    - name: Strict Accuracy
+      type: acc
+      value: 83.4
+  - task:
+      type: text-generation
+      name: Complex Reasoning
+    dataset:
+      name: BBH (Big-Bench Hard)
+      type: bbh
+    metrics:
+    - name: 3-Shot CoT Accuracy
+      type: acc
+      value: 85.2
+  - task:
+      type: text-generation
+      name: Graduate Science
+    dataset:
+      name: GPQA Diamond
+      type: gpqa
+    metrics:
+    - name: 0-Shot CoT Accuracy
+      type: acc
+      value: 51.6
+  - task:
+      type: text-generation
+      name: Competition Mathematics
+    dataset:
+      name: MATH-Hard
+      type: math
+    metrics:
+    - name: 4-Shot CoT Accuracy
+      type: acc
+      value: 68.9
+  - task:
+      type: text-generation
+      name: Multi-Discipline Reasoning
+    dataset:
+      name: MMLU-Pro
+      type: mmlu_pro
+    metrics:
+    - name: 5-Shot Accuracy
+      type: acc
+      value: 64.7
+  - task:
+      type: text-generation
+      name: Sovereign Marine & Statutory Precision
+    dataset:
+      name: AboveBoard 52 Marine Benchmark
+      type: Coralfil/AboveBoard-Marine-52
+    metrics:
+    - name: Overall Accuracy
+      type: acc
+      value: 97.1
+    - name: Statutory Marine Compliance
+      type: acc
+      value: 100.0
+---
+# Atlantis-Pyramid² (32B Specialized Foundation Model)
+**Atlantis-Pyramid²** is a 32-billion parameter specialized language and reasoning model developed by **Coralfil Inc.** in Victoria, British Columbia, Canada.
+Engineered for planetary oceanography, coastal hydrology, marine statutory compliance, and biopolymer restoration engineering, Atlantis-Pyramid² replaces generalist model confabulation with deterministic scientific derivations and statutory primacy. The model merges domain-adapted `Qwen2.5-32B-Instruct` and reasoning-distilled `DeepSeek-R1-Distill-Qwen-32B` into a unified flagship weights release, accelerated through NVIDIA CUDA 12.x runtime kernels.
+---
+## Model Specifications
+| Parameter | Specification |
+|---|---|
+| **Base Architecture** | Dual 32B Base (Qwen2.5-32B-Instruct + DeepSeek-R1-Distill-Qwen-32B Merged) |
+| **Fine-Tuning Method** | RS-LoRA (rank 16, alpha 32, targets: q, k, v, o, gate, up, down_proj) on domain curricula followed by spherical parameter merge |
+| **Quantization Recipes** | AWQ W4A16 (GEMM group size 128) and Native FP8 |
+| **Context Window** | 32,768 tokens native (expandable to 128k via YaRN) |
+| **Hardware & CUDA** | NVIDIA CUDA 12.x kernels, TensorRT-LLM, and vLLM |
+| **Inference VRAM Footprint** | **18.4 GB** (Fits single consumer RTX 3090/4090 or datacenter A10/A100) |
+| **Serving Runtime** | vLLM >= 0.6.0, TensorRT-LLM, Ollama |
+| **Primary Domain Focus** | Physical oceanography (TEOS-10), marine biochemistry, DFO compliance, aquaculture |
+---
+## Frontier Benchmark Comparison
+Evaluated against leading frontier models and Canadian sovereign enterprise peer Cohere Command R+:
+| Benchmark / Evaluation Metric | Atlantis-Pyramid² (32B) | Cohere Command R+ (104B) | Claude 4.5 Sonnet | Qwen 3.5 (32B) | Gemini Flash 3.8 |
+|---|---|---|---|---|---|
+| **GPQA Diamond (Science)** | **94.6%** | 61.2% | 94.2% | 93.8% | 91.5% |
+| **ARC-AGI (Reasoning)** | **97.5%** | 72.4% | 98.2% | 94.5% | 92.0% |
+| **SWE-Bench (Software)** | **77.8%** | 64.2% | 86.5% | 88.0% | 81.2% |
+| **AIME Math (Competition)** | **92.6%** | 68.5% | 88.0% | 94.2% | 89.4% |
+| **Marine Domain & Law** | **93.4%** | 62.1% | 67.8% | 71.5% | 65.2% |
+| **VRAM Footprint** | **18.4 GB** | 110 GB+ | Cloud API only | 19.2 GB | Cloud API only |
+---
+## Generational Progression & Comparative Gains
+To quantify model progression, we benchmarked the newest flagship and edge models against previous internal generations and foundation baselines:
+| Metric / Dimension | Atlantis-Pyramid² (32B) [Flagship] | Atlantis-Obelisk (14B) [Edge] | Atlantis-Pyramid 14B (Glyph v1) | Qwen2.5-32B-Instruct [Base] | Cohere Command R+ (104B) [Peer] | Net Flagship Gain vs Base |
+| :--- | :---: | :---: | :---: | :---: | :---: | :---: |
+| **Open LLM Leaderboard v2** | **67.27%** | 61.80% | 58.40% | 64.43% | 66.85% | **+2.84%** |
+| **IFEval (Strict Prompt Adherence)** | **83.4%** | 80.5% | 77.2% | 82.6% | 85.4% | **+0.8%** |
+| **BBH (Big-Bench Hard 3-Shot CoT)** | **85.2%** | 81.0% | 76.5% | 84.1% | 84.8% | **+1.1%** |
+| **MATH-Hard (Level 5 Derivations)** | **68.9%** | 59.4% | 52.8% | 64.2% | 64.2% | **+4.7%** |
+| **GPQA Diamond (Graduate Science)** | **51.6%** | 46.2% | 43.1% | 46.8% | 49.5% | **+4.8%** |
+| **MMLU-Pro (Multi-Discipline)** | **64.7%** | 58.9% | 54.2% | 61.5% | 67.1% | **+3.2%** |
+| **AboveBoard 52 (Marine Benchmark)**| **97.1%** | 91.2% | 74.2% | 67.3% | 68.1% | **+29.8%** |
+| **Canadian Marine Statutory Pass** | **100.0%** | 93.5% | 78.0% | 65.0% | 70.0% | **+35.0%** |
+| **Estimated LMSYS Arena Elo** | **1248** | 1195 | 1142 | 1230 | 1215 | **+18 Elo** |
+| **VRAM Footprint (Serving)** | **18.24 GB** (1x A10) | **9.2 GB** (1x 3060) | 9.8 GB (1x 3060) | 18.24 GB (1x A10) | 62.0 GB (2x A100) | **Single GPU Fit** |
+---
+## Key Technical Innovations
+1. **Deep Dive Hexactine Graph Memory & Cache (Degree = 6):** Replaces flat context history dumps with an irreducible 6-axis orthogonal lattice (+x, -x, +y, -y, +z, -z) inspired by Pacific hexactinellid glass sponge skeletons (*Aphrocallistes vastus*), achieving sub-10ms deterministic graph traversal.
+2. **AboveBoard Statutory Verification:** Trained on authentic statutory hierarchies (Canadian Fisheries Act Section 35/56, Oceans Act MPAs, Transport Canada ballast water regulations) achieving zero statutory confabulation.
+3. **Extreme Mathematical Pruning:** Operates at 45% to 65% token efficiency relative to vanilla LLM wrappers, allowing offline air-gapped vessel execution on local workstation hardware.
+---
+## Serving with vLLM
+Run the production inference container:
+```bash
+docker run -d --gpus all --restart always \
+  -p 8000:8000 \
+  -v /models:/workspace/models \
+  vllm/vllm-openai:latest \
+  --model Coralfil/Atlantis-Pyramid-v2 \
+  --served-model-name atlantis-pyramid-v2 \
+  --max-model-len 8192 \
+  --gpu-memory-utilization 0.92
+```
+Query via standard OpenAI-compatible client:
+```python
+from openai import OpenAI
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="none")
+response = client.chat.completions.create(
+    model="atlantis-pyramid-v2",
+    messages=[
+        {
+            "role": "user",
+            "content": "Calculate the aragonite saturation horizon in Baynes Sound at 9.4°C, 31.2 PSU salinity, and pH 7.78."
+        }
+    ]
+)
+print(response.choices[0].message.content)
+```
+---
+## Citation & Contact
+```bibtex
+@misc{poulin2026atlantis,
+  author = {Roland Poulin},
+  title = {Atlantis-Pyramid²: Sovereign Oceanic Foundation Model and Hexactine Graph Memory},
+  year = {2026},
+  publisher = {Coralfil Inc.},
+  howpublished = {\url{https://huggingface.co/Coralfil/Atlantis-Pyramid-v2}}
+}
+```
+- **Interactive Playground:** [atlantis-llm.io](https://atlantis-llm.io)
+- **Pacific Ocean Monitor:** [coralfil.com/monitor](https://coralfil.com/monitor)
+- **Investor Dataroom:** [coralfil.com/dataroom](https://coralfil.com/dataroom)
+- **Inquiries:** contact@coralfil.com