Qwable PICT
Work-in-Progress / Experimental Test Release
An early exploration of porting the Qwable-EX behavioral and capability profile onto the successor / updated architecture of ByteMythos.
Overview
Qwable PICT is a proof-of-concept model that tests whether the reasoning style, tool-use patterns, and overall character of charlieduzstuf/Qwable-Ex can be successfully transferred onto the next-generation architecture developed from SAXON-AI/ByteMythos.
This is not a finished production model. It is an architecture migration experiment intended to:
- Validate compatibility between the Qwable-EX training lineage and the updated ByteMythos architecture
- Surface any architectural mismatches, capability regressions, or unexpected strengths early
- Serve as a baseline for future, more complete ports and fine-tunes
Expect incomplete features, uneven performance, and rapid iteration.
Motivation
Qwable-EX demonstrated a distinctive combination of structured reasoning and agentic behavior. ByteMythos (and its successors) introduced architectural changes aimed at improved efficiency, byte-level or more granular modeling characteristics, and better scaling properties.
Qwable PICT is the first public test of marrying those two lines of work: taking the behavioral and capability characteristics of Qwable-EX and placing them on the updated architectural substrate of the ByteMythos family.
Status
| Aspect | Current State |
|---|---|
| Architecture port | Experimental / incomplete |
| Training | Limited / test-scale |
| Evaluation | Informal only |
| Intended use | Research, architecture validation, feedback |
| Production readiness | None |
Base Models
- Behavioral / capability source:
charlieduzstuf/Qwable-Ex - Architectural substrate:
SAXON-AI/ByteMythos(successor / updated version)
Intended Use & Limitations
Intended for:
- Researchers and developers interested in architecture migration experiments
- Early testing of whether Qwable-style behavior survives the move to the newer ByteMythos-derived architecture
- Feedback and bug reports that can guide subsequent iterations
Not intended for:
- Production deployments
- Safety-critical or high-stakes applications
- Benchmark claims or competitive comparisons (the model is still too early-stage)
Known limitations (expected at this stage):
- Incomplete transfer of Qwable-EX capabilities
- Possible degradation in coherence, tool-use reliability, or reasoning depth
- Untested edge cases and long-context behavior
- Rapid changes between versions; this card may lag behind the actual weights
License
Microsoft Public License (MS-PL)
Acknowledgements
- Original Qwable-EX work by the charlieduzstuf lineage
- ByteMythos architecture by SAXON-AI and subsequent updates
This is an early test artifact. Feedback, observations, and failure cases are welcome and will help shape future versions.