Spaces:
Paused
Paused
Commit History
-
Report throughput in consistent units
tok/s and s/token are reciprocals; quoting one for each side reads as a comparison but is not one. Both rows now give both: 1.43 vs 3.02 tok/s.b582a21 -
Correct the latency claim: measure against llama.cpp, not our own worst case
llama-bench on this machine, CPU-only, same model family: 3.02 tok/s (0.33 s/token) Q4_K_M, against ~0.7 s/token here. We are 2.1x SLOWER than the reference implementation. The previous '68 s -> 0.7 s' framing used an unoptimised numpy path as the baseline, which is our own worst case and not a referenceable measurement.82bf4d8 -
Correct README to the verified claims
Was: Gemma-4-12B ternary, 48 layers, 'no matmul'. Is: Llama-3.2-3B int8 per-column, 28 layers, and it performs matmuls -- the wave form is an equivalence proof about the arithmetic, not a replacement for it. Adds the measured numbers, the ternary result on both model families, and the DAG/bandwidth analysis.65d070e
-
Filter empty thoughts from recording and chat display
Verified signaturee431d79 -
Collapse empty chat-log black box; chat column hugs text bar
Verified signaturecdea7ed -
Full-vocab HVM cache (128K tokens) via chunked bf16 projection
Verified signaturef7a2fac -
Thought recording + contemplate mode + denoise parser for int8 digit collapse
Verified signature3f49d51 -
fix: chat messages persist + agent generates 256 tokens per round (was 32)
Verified signaturecf620e1 -
Trim test target to Space-shipping files (C libs + py_compile only)
Verified signature6cce3a6 -
Drop __pycache__, clean slate
Verified signature9e19cd4 -
Port full dashboard engine: phoenix_dashboard + agent_core + hyper_vocab_memory + bqsm_assist
Verified signaturebdf5f4a
-
Add chat debug logging
Nicholas Bumgarnerc1298e1 -
Force rebuild
Nicholas Bumgarner92af58e -
Increase chat timeout to 120s for Space CPU
Nicholas Bumgarner933ff98 -
Init ring buffers in start_mixer
Nicholas Bumgarnera570df2 -
Support both phoenix and mixer engines in chat
Nicholas Bumgarner8f7952f -
Update mixer.c with LM head projection + ring buffer chat I/O
Nicholas Bumgarnerb918c0d -
Update mixer.c with actual ternary weight decoding from model
Nicholas Bumgarner2ebca3e -
Add mixer.c source to Space repo
Nicholas Bumgarner18881f2 -
Add Start Mixer button + start_mixer function for fast mixing ring engine
Nicholas Bumgarnercdad13a -
Revert to ring buffer protocol; optimize log tail reading
Nicholas Bumgarnerb356ed7 -
Fix chat protocol: use named pipe (/tmp/phoenix_chat.pipe) per phoenix_brain.c chat mode
Nicholas Bumgarner76b2b96 -
Download binary + model to /tmp to avoid Space auto-restart
Nicholas Bumgarnerec34b87 -
Cache binary download (no force_download); fix indentation
Nicholas Bumgarner3aec1bb -
Fix stop_phoenix: minimize lock holding, close fds outside lock
Nicholas Bumgarnera53b1da -
Add phoenix daemon monitor thread + tick log display
Nicholas Bumgarner488f97e -
Force re-download binary; fix lock contention in get_boot_status
Nicholas Bumgarnerfaa1f0b -
Fix lock contention: get_boot_status reads without lock, phox_chat checks minimal
Nicholas Bumgarner3e5b14e -
Download pre-built phoenix binary from HF Hub instead of compiling on Space
Nicholas Bumgarneradcbc7e -
Pre-built static phoenix binary; skip compilation on boot
Nicholas Bumgarnerd0f24f6 -
Manual boot: tokenizer loads at startup, full boot on button click. Non-blocking stop_phoenix.
Nicholas Bumgarner4eb0c7c -
Remove blocking wait loop; tick() checks daemon log instead
Nicholas Bumgarner9317c1d -
Non-blocking boot wait: 0.5s sleep, only last 20 log lines
Nicholas Bumgarnerfdff78d -
Fix boot wait loop: release lock between checks, extended timeout
Nicholas Bumgarnere73f64f -
Touch app.py to force Space rebuild
Nicholas Bumgarner6f7f402 -
Fix tokenizer get_vocab API call
Nicholas Bumgarner0ace216 -
Fix Gradio 6.22 + remove torch dep: use tokenizers lib, gr.Timer API
Nicholas Bumgarner4aa8326 -
Fix Gradio 6.0 compat: remove css/interactive params, fix tokenizer loading
Nicholas Bumgarner26b395e -
Fix short_description length
Nicholas Bumgarnerfb8c492 -
Convert to Gradio: Phox wave-rider inference engine
Nicholas Bumgarner- Replaced static HTML with Gradio chat interface - Added phoenix_brain.c (BQSM v8 harmonic inference engine) - Space downloads model from compunerd/emerging-systems-models - Compiles C binary at startup with AVX2 + OpenMP - Shared-memory ring buffer I/O between Python and C - Tokenizes via Gemma 4 tokenizer from HF Huba2124e2
-
initial commit
Verified signature850c7dc