license: apache-2.0
base_model: saidutta69/MiniCPM5-2B-Supertool
library_name: transformers
pipeline_tag: text-generation
tags:
- tool-calling
- function-calling
- agents
- qlora
- claude-code
- heretic
- uncensored
- decensored
- abliterated
- reproducible
datasets: - saidutta69/fable-5.1-premium-v2
- MoreThought/Fable-5.1-Max-Reasoning-Filtered-10000x
language: - en
This is a decensored version of saidutta69/MiniCPM5-2B-Supertool, made using Heretic v2.1.0
[!TIP]
This model is reproducible!See the README in the
reproducedirectory for more information.
Arbitrary-Rank Ablation (ARA) parameters
| Parameter | Value |
|---|---|
| start_layer_index | 19 |
| end_layer_index | 35 |
| preserve_good_behavior_weight | 0.7948 |
| steer_bad_behavior_weight | 0.0102 |
| overcorrect_relative_weight | 0.7500 |
| neighbor_count | 8 |
Performance
| Metric | This model | Original model (saidutta69/MiniCPM5-2B-Supertool) |
|---|---|---|
| Refusals (Classifier) | 12/100 | 96/100 |
| KL divergence | 0.0494 | 0 (by definition) |
🛠️ MiniCPM5-2B-Supertool
A 2.5B tool-calling model: openbmb/MiniCPM5-2B fine-tuned to drive the Claude Code
tool surface — Bash, Read, Write, Edit, Glob, Grep, WebFetch,WebSearch, TodoWrite, TaskUpdate, ToolSearch — emitting calls in the base
model's native XML form:
<function name="Bash"><param name="command">ls tasks/</param></function>
Weights are full precision with the LoRA already merged. Load it directly:
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("saidutta69/MiniCPM5-2B-Supertool")
model = AutoModelForCausalLM.from_pretrained("saidutta69/MiniCPM5-2B-Supertool")
No PEFT, no adapter step. Pass your tool schemas to apply_chat_template(tools=...)
as you would with the base model.
Benchmarks
FC-Eval — independent, held-out APIs
saidutta69/fc-eval-benchmark,
eval split: 89 APIs, none seen in training. 150 tasks sampled round-robin across
APIs, greedy decoding, both models at fp16. This is the one result here that does not
come from the training corpus.
| Metric | Base | Supertool |
|---|---|---|
| Task success (exact call) | 49.3% | 77.3% |
| Tool name correct | 50.0% | 90.0% |
| Argument keys correct | 50.0% | 87.3% |
| Format failures | 50.0% | 8.7% |
| Wall time, 150 tasks | 1071 s | 886 s |
The base is all-or-nothing: it emits a parseable call exactly half the time, and when
it does the call is exactly right (success == name == arg-keys == 0.50). Supertool
parses 91.3% of the time and is exact on 77.3%. Most of the gain is format compliance
and tool choice; argument values improve less.
In-domain agent traces
120 examples from the training dataset's own validation split, never trained on.
Greedy, max_new=384, both models 4-bit on one Tesla T4.
| Metric | Base | Supertool |
|---|---|---|
| Completion loss | 1.5691 | 1.1246 |
| Tool name correct | 47.5% | 54.2% |
| Argument names correct | 31.7% | 46.7% |
| Full match | 0.0% | 4.2% |
| Parseable call emitted | 88.3% | 75.8% |
That last row looks like a regression and is not one. Every failing Supertool example
is a Write call cut off inside its content argument by the generation cap: this
model writes an actual file body, while the base emits a shorter, vaguer call that
happens to fit. Raising the cap from 160 to 384 lifted the base from 81.7% to 88.3%
while Supertool moved 64.2% to 75.8% — truncation, not non-compliance. Among calls it
does emit:
| Base | Supertool | |
|---|---|---|
| Tool name correct | 53.8% | 71.4% |
| Argument names correct | 35.8% | 61.5% |
Training
| Base | openbmb/MiniCPM5-2B |
| Method | QLoRA — NF4 4-bit, LoRA r=16 alpha=32 dropout 0 |
| LoRA targets | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
| Data | saidutta69/fable-5.1-premium-v2, claude_code split only |
| Examples | 42,514 — one per assistant tool-calling turn |
| Coverage | ~3,500 of 4,132 trajectories, 12 examples each |
| Validated calls | 347,136 — zero bad JSON, zero missing required fields |
| Supervision | final assistant turn only, via a prompt/completion split |
| Schedule | 240 steps, effective batch 16, seq 4096, lr 2e-4 cosine, warmup 3% |
| Loss | 1.107 → 0.276 |
| Hardware | 2× Tesla T4 (sm_75), fp16, 7 h 40 m, peak 8.92 GB per card |
Data
Traces originate fromMoreThought/Fable-5.1-Max-Reasoning-Filtered-10000x;
credit for them belongs to MoreThought. They were repackaged intosaidutta69/fable-5.1-premium-v2
and then adapted for this model:
- kept only the
claude_codeshare; dropped the anonymised half (unknown_tool) - dropped
tool_callswith no matching tool response, and orphantoolmessages - re-serialised messages canonically and de-duplicated by SHA-256 over messages plus
tool names - reconstructed the 14-tool Claude Code schema surface, validated every call against
it, dropped rows that did not fit - excluded the
fable_codeconfiguration: 203,750 of its 203,794 calls carry the
schema rendered as the value ({"path": "<file_path>"}), which teaches a model to
emit placeholders - sampled 12 examples per trajectory so coverage spans ~3,500 rows instead of 212
Limitations
- Not a general assistant. A 2.5B specialist that emits tool calls; it is weaker
than the base at open-ended conversation. - One tool family. Trained on Claude Code's surface, not arbitrary schemas. The
FC-Eval result shows real cross-schema transfer, not generalisation to any API. - Exact match is low in-domain — 5 of 120. Long
Writepayloads are approximated. - One pass, never looped. 240 steps is roughly three quarters of a single epoch.
- No GGUF / llama.cpp build published yet.
- Held-out loss is worse than an earlier narrower run of the same recipe
(1.125 vs 0.976) while tool accuracy is better: wider data memorises less and
generalises more.
Responsible use
No guardrails. This model was not trained or evaluated for prompt-injection
resistance, and it will follow instructions that arrive inside tool output. Validate
its calls before letting it act on anything destructive, and do not expose it to
untrusted tool results without a policy layer in front.
Licence
Apache-2.0.
openbmb/MiniCPM5-2B states in its card that the repository and MiniCPM model weights
are Apache-2.0, with a link to the LICENSE in the OpenBMB GitHub repository. The
training data declares Apache-2.0 in saidutta69/fable-5.1-premium-v2, inherited from
MoreThought, which declares the same.
Both declarations are made in cards: neither dataset repository ships a standaloneLICENSE file, and the MoreThought source carries the tag with no separate terms or
attribution section. The Apache-2.0 claim for these weights therefore rests on those
upstream declarations. Stated plainly here rather than implied.
Apache-2.0 requires keeping the licence and notices and indicating changes — the
change list above is provided for that purpose.
🔗 Related
- saidutta69/fable-5.1-premium-v2 — training data (
claude_codesplit) - saidutta69/fc-eval-benchmark — the independent benchmark above
- MoreThought/Fable-5.1-Max-Reasoning-Filtered-10000x — upstream traces
- RACER IS OP — SFT & Distillation — collection
Made with care by RACER IS OP (saidutta69).