Text Generation
Transformers
English
custom
tokenizer
symbolic-ai
mathematics
llm
reasoning
ast
compiler
nlp
deep-learning
machine-learning
mathematical-reasoning
symbolic-reasoning
tokenization
parser
artificial-intelligence
Eval Results (legacy)
Instructions to use Eduloop/mathtok with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Eduloop/mathtok with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Eduloop/mathtok")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("Eduloop/mathtok", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Eduloop/mathtok with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Eduloop/mathtok" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Eduloop/mathtok", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/Eduloop/mathtok
- SGLang
How to use Eduloop/mathtok with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "Eduloop/mathtok" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Eduloop/mathtok", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "Eduloop/mathtok" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Eduloop/mathtok", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use Eduloop/mathtok with Docker Model Runner:
docker model run hf.co/Eduloop/mathtok
| <svg width="100%" viewBox="0 0 680 820" role="img" style="" xmlns="http://www.w3.org/2000/svg"> | |
| <title style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto">MathTok 7-layer pipeline with proposed improvements per layer</title> | |
| <desc style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto">Structural diagram showing each of the 7 layers of the MathTok pipeline, color-coded by status: teal for strong layers, amber for layers needing improvement, and red for critical gaps.</desc> | |
| <defs> | |
| <marker id="arrow" viewBox="0 0 10 10" refX="8" refY="5" markerWidth="6" markerHeight="6" orient="auto-start-reverse"> | |
| <path d="M2 1L8 5L2 9" fill="none" stroke="context-stroke" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round"/> | |
| </marker> | |
| </defs> | |
| <text x="40" y="28" style="font-size:11px;fill:#1D9E75;font-weight:500;;fill:rgb(29, 158, 117);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:11px;font-weight:500;text-anchor:start;dominant-baseline:auto">SOLID</text> | |
| <rect x="80" y="18" width="10" height="10" rx="2" fill="#1D9E75" style="fill:rgb(29, 158, 117);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="100" y="28" style="font-size:11px;fill:#BA7517;font-weight:500;;fill:rgb(186, 117, 23);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:11px;font-weight:500;text-anchor:start;dominant-baseline:auto">NEEDS WORK</text> | |
| <rect x="172" y="18" width="10" height="10" rx="2" fill="#BA7517" style="fill:rgb(186, 117, 23);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="192" y="28" style="font-size:11px;fill:#A32D2D;font-weight:500;;fill:rgb(163, 45, 45);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:11px;font-weight:500;text-anchor:start;dominant-baseline:auto">CRITICAL GAP</text> | |
| <rect x="270" y="18" width="10" height="10" rx="2" fill="#A32D2D" style="fill:rgb(163, 45, 45);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <g onclick="sendPrompt('How should the canonicalization engine handle unsimplifiable expressions?')" style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"> | |
| <rect x="40" y="46" width="280" height="56" rx="8" stroke-width="0.5" style="fill:rgb(225, 245, 238);stroke:rgb(15, 110, 86);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="56" y="68" style="fill:rgb(8, 80, 65);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">L1 — Canonicalization</text> | |
| <text x="56" y="86" style="fill:rgb(15, 110, 86);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">Add timeout guard; cache sympy results</text> | |
| </g> | |
| <text x="334" y="66" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Timeout for sympy hangs on complex</text> | |
| <text x="334" y="80" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto"> exprs (e.g. `((x+1)²+(y-1)²)³`)</text> | |
| <text x="334" y="94" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ LRU cache to avoid re-parsing duplicates</text> | |
| <line x1="180" y1="102" x2="180" y2="120" stroke="#1D9E75" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <g onclick="sendPrompt('How can the hybrid lexer better handle mixed text and math?')" style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"> | |
| <rect x="40" y="120" width="280" height="56" rx="8" stroke-width="0.5" style="fill:rgb(252, 235, 235);stroke:rgb(163, 45, 45);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="56" y="142" style="fill:rgb(121, 31, 31);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">L2 — Hybrid lexer</text> | |
| <text x="56" y="160" style="fill:rgb(163, 45, 45);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">Critical: improve text+math span detection</text> | |
| </g> | |
| <text x="334" y="140" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Confidence score per span (math vs text)</text> | |
| <text x="334" y="154" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Handle inline math without delimiters</text> | |
| <text x="334" y="168" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Fallback to GPT-2 BPE for ambiguous spans</text> | |
| <line x1="180" y1="176" x2="180" y2="194" stroke="#E24B4A" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <g onclick="sendPrompt('What improvements can be made to the AST generator?')" style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"> | |
| <rect x="40" y="194" width="280" height="56" rx="8" stroke-width="0.5" style="fill:rgb(225, 245, 238);stroke:rgb(15, 110, 86);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="56" y="216" style="fill:rgb(8, 80, 65);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">L3 — AST generator</text> | |
| <text x="56" y="234" style="fill:rgb(15, 110, 86);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">Add parse failure fallback; depth cap</text> | |
| </g> | |
| <text x="334" y="214" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Max depth cap (avoids 145-token explosions)</text> | |
| <text x="334" y="228" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Graceful fallback when sympy parse fails</text> | |
| <text x="334" y="242" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Track parse confidence per subtree</text> | |
| <line x1="180" y1="250" x2="180" y2="268" stroke="#1D9E75" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <g onclick="sendPrompt('How can the semantic registry be extended for more math domains?')" style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"> | |
| <rect x="40" y="268" width="280" height="56" rx="8" stroke-width="0.5" style="fill:rgb(250, 238, 218);stroke:rgb(133, 79, 11);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="56" y="290" style="fill:rgb(99, 56, 6);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">L4 — Semantic registry</text> | |
| <text x="56" y="308" style="fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">Extend to geometry, stats, set theory</text> | |
| </g> | |
| <text x="334" y="288" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Domain tags: geometry, stats, logic</text> | |
| <text x="334" y="302" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Commutativity flag (x+y ≡ y+x at token level)</text> | |
| <text x="334" y="316" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Inverse-pair registry (sin/arcsin, e^x/ln)</text> | |
| <line x1="180" y1="324" x2="180" y2="342" stroke="#EF9F27" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <g onclick="sendPrompt('What are the tradeoffs of DFS preorder vs other tree serialization orders?')" style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"> | |
| <rect x="40" y="342" width="280" height="56" rx="8" stroke-width="0.5" style="fill:rgb(250, 238, 218);stroke:rgb(133, 79, 11);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="56" y="364" style="fill:rgb(99, 56, 6);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">L5 — Structural serializer</text> | |
| <text x="56" y="382" style="fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">Add bracket tokens; test alt. traversals</text> | |
| </g> | |
| <text x="334" y="362" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Explicit scope tokens [SCOPE_OPEN/CLOSE]</text> | |
| <text x="334" y="376" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Experiment with postorder for calculus ops</text> | |
| <text x="334" y="390" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Subtree hash for dedup at sequence level</text> | |
| <line x1="180" y1="398" x2="180" y2="416" stroke="#EF9F27" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <g onclick="sendPrompt('How should structural attention metadata be validated experimentally?')" style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"> | |
| <rect x="40" y="416" width="280" height="56" rx="8" stroke-width="0.5" style="fill:rgb(252, 235, 235);stroke:rgb(163, 45, 45);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="56" y="438" style="fill:rgb(121, 31, 31);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">L6 — Attention metadata</text> | |
| <text x="56" y="456" style="fill:rgb(163, 45, 45);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">Critical: validate metadata improves training</text> | |
| </g> | |
| <text x="334" y="436" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Ablation study: does metadata help accuracy?</text> | |
| <text x="334" y="450" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Add sibling-count and parent-op fields</text> | |
| <text x="334" y="464" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Expose metadata as optional (not forced on)</text> | |
| <line x1="180" y1="472" x2="180" y2="490" stroke="#E24B4A" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <g onclick="sendPrompt('How should the BPE vocabulary handle unknown math symbols?')" style="fill:rgb(0, 0, 0);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"> | |
| <rect x="40" y="490" width="280" height="56" rx="8" stroke-width="0.5" style="fill:rgb(250, 238, 218);stroke:rgb(133, 79, 11);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="56" y="512" style="fill:rgb(99, 56, 6);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">L7 — Two-tier vocabulary</text> | |
| <text x="56" y="530" style="fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">Add unknown-symbol handling; grow vocab</text> | |
| </g> | |
| <text x="334" y="510" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ [UNK_MATH] token for unseen symbols</text> | |
| <text x="334" y="524" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Grow fixed vocab to cover Greek/set notation</text> | |
| <text x="334" y="538" style="fill:rgb(61, 61, 58);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:start;dominant-baseline:auto">+ Verify HF tokenizer fast-path is preserved</text> | |
| <line x1="180" y1="546" x2="180" y2="564" stroke="#EF9F27" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <rect x="40" y="564" width="280" height="44" rx="8" fill="#E1F5EE" stroke="#0F6E56" stroke-width="0.5" style="fill:rgb(225, 245, 238);stroke:rgb(15, 110, 86);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="180" y="582" text-anchor="middle" style="fill:#085041;;fill:rgb(8, 80, 65);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:middle;dominant-baseline:auto">Output: structured token stream</text> | |
| <text x="180" y="597" text-anchor="middle" style="fill:#0F6E56;;fill:rgb(15, 110, 86);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">IDs + AST metadata + BPE fallback</text> | |
| <line x1="180" y1="608" x2="180" y2="626" stroke="#1D9E75" marker-end="url(#arrow)" style="fill:none;stroke:rgb(115, 114, 108);color:rgb(0, 0, 0);stroke-width:1.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <rect x="40" y="626" width="600" height="148" rx="10" fill="#FAEEDA" stroke="#BA7517" stroke-width="0.5" style="fill:rgb(250, 238, 218);stroke:rgb(186, 117, 23);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="60" y="651" style="fill:#633806;;fill:rgb(99, 56, 6);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:14px;font-weight:500;text-anchor:start;dominant-baseline:auto">3 architecture-level proposals (cross-cutting)</text> | |
| <rect x="56" y="662" width="174" height="96" rx="6" fill="#fff" stroke="#EF9F27" stroke-width="0.5" style="fill:rgb(255, 255, 255);stroke:rgb(239, 159, 39);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="143" y="681" text-anchor="middle" style="fill:#412402;font-size:12px;;fill:rgb(65, 36, 2);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:500;text-anchor:middle;dominant-baseline:auto">Streaming tokenizer</text> | |
| <text x="143" y="697" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">Process input token-by-token,</text> | |
| <text x="143" y="711" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">not all at once. Enables</text> | |
| <text x="143" y="725" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">real-time LLM generation.</text> | |
| <text x="143" y="739" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">Critical for inference.</text> | |
| <rect x="253" y="662" width="174" height="96" rx="6" fill="#fff" stroke="#EF9F27" stroke-width="0.5" style="fill:rgb(255, 255, 255);stroke:rgb(239, 159, 39);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="340" y="681" text-anchor="middle" style="fill:#412402;font-size:12px;;fill:rgb(65, 36, 2);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:500;text-anchor:middle;dominant-baseline:auto">Round-trip validator</text> | |
| <text x="340" y="697" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">Decode tokens back to</text> | |
| <text x="340" y="711" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">expression, verify it matches</text> | |
| <text x="340" y="725" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">original via sympy.equals().</text> | |
| <text x="340" y="739" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">Catches silent loss of meaning.</text> | |
| <rect x="450" y="662" width="174" height="96" rx="6" fill="#fff" stroke="#EF9F27" stroke-width="0.5" style="fill:rgb(255, 255, 255);stroke:rgb(239, 159, 39);color:rgb(0, 0, 0);stroke-width:0.5px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:16px;font-weight:400;text-anchor:start;dominant-baseline:auto"/> | |
| <text x="537" y="681" text-anchor="middle" style="fill:#412402;font-size:12px;;fill:rgb(65, 36, 2);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:500;text-anchor:middle;dominant-baseline:auto">Benchmark suite</text> | |
| <text x="537" y="697" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">Expand test set beyond 63</text> | |
| <text x="537" y="711" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">expressions. Add ODE, PDE,</text> | |
| <text x="537" y="725" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">matrix, tensor, probability</text> | |
| <text x="537" y="739" text-anchor="middle" style="fill:#854F0B;;fill:rgb(133, 79, 11);stroke:none;color:rgb(0, 0, 0);stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;opacity:1;font-family:"Anthropic Sans", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;font-size:12px;font-weight:400;text-anchor:middle;dominant-baseline:auto">notation categories.</text> | |
| </svg> |