Falcon-H1 TinyShell
Fine-tuned compact language model for the TinyShell ShellIntent
natural-language-to-structured-IR task.
Base model
tiiuae/Falcon-H1-Tiny-Tool-Calling-90M
This is a supervised fine-tune of the upstream Falcon-H1 Tiny tool-calling
model. The derivative is distributed under the upstream
Falcon-LLM License;
see
LICENSE for the authoritative license link.
Task
The model converts natural-language instructions into structured TinyShell
ShellIntent JSON.
Supported high-level decisions include:
compileclarifyunsupported
Held-out evaluation
| Metric | Result |
|---|
| JSON parse rate | 99.50% |
| Schema validity | 96.50% |
| IR exact match | 33.00% |
| Decision accuracy | 95.50% |
| Operation accuracy | 58.00% |
| Slot precision | 65.37% |
| Slot recall | 57.80% |
| Slot F1 | 61.35% |
| Risk accuracy | 94.00% |
| Confirmation accuracy | 95.00% |
| Clarify accuracy | 100.00% |
| Unsupported accuracy | 30.00% |
| Multi-operation accuracy | 22.22% |
| Median inference latency | 3700.2038380005615 ms |
Generation policy
JSON-completion stopping criterion
FunctionGemma and Falcon-H1 initially produced a valid first JSON object but
frequently continued generating additional content. Their corrected final
evaluation uses a generation-time stopping criterion that terminates once the
first complete top-level JSON object is generated. This is generation control,
not post-hoc JSON repair.
LFM2.5 terminated correctly under the original inference configuration.
Training
The model was fine-tuned with supervised causal language modeling.
- Seed:
42
- Best validation loss:
0.11050145771034295
- Training time:
2314.405442636 seconds
- Peak GPU memory:
2.0683422088623047 GB
Prompt tokens were masked from the language-model loss and the assistant JSON
response was used as the supervised target.
Training used 1,600 examples, with 200 validation examples and 200 held-out
test examples. The random seed was 42. The frozen source hashes and complete
training metadata are included in evaluation/training_result.json.
Included files
- Fine-tuned model weights
- Model configuration
- Tokenizer / processor files
- Chat template when saved
- Generation configuration when saved
evaluation/final_metrics.jsonevaluation/test_predictions.jsonlevaluation/training_result.jsoninference_example.pyrequirements.txtLICENSE and NOTICESHA256SUMS.txt
Limitations
This pilot used one training seed. Test-set bootstrap intervals quantify
held-out sample uncertainty but do not replace independent repeated training.
Exact ShellIntent matching is intentionally strict: one incorrect operation,
argument, or structured field makes the complete IR prediction incorrect.
This model emits untrusted structured intent. Do not execute model output
directly. Validate the JSON against the TinyShell schema, compile it through a
deterministic platform-aware compiler, apply safety checks, and require user
confirmation where appropriate.
License
The model weights are a derivative of
tiiuae/Falcon-H1-Tiny-Tool-Calling-90M and are subject to the
Falcon-LLM License.
The license link is also recorded in
LICENSE.
The TinyShell training data contribution is attributed under
CC BY
4.0. Upstream source material
may have separate terms; see the TinyShell dataset documentation for details.
