A QLoRA fine-tune of Empero/Qwen3.8-9B (Qwen3.5 hybrid-architecture distillation, 9B dense, 262,144-token context). Trained on real Hermes agent trajectories, not synthetic data.
<|tool_call|>, <|tool_response|>)Read this before you use the numbers below.
After this checkpoint was published, we tested the base model
(Empero/Qwen3.8-9B-Distill, no fine-tuning) through the live Hermes
interface: the OpenAI-native tools parameter path that Hermes actually
uses. The base model calls tools correctly on its own.
The harness that certified this checkpoint did not use that interface.
It baked tool schemas into the system text, never passed tools, and
required raw <tool_call> XML in the response. That prompt shape matches
neither the base model's own chat template nor the live Hermes request
path. Every gate score in the v26-to-v29 lineage measured this synthetic
interface.
The statement in this README that "the base distill contributes zero tool-calling" is incorrect. It measured the synthetic interface.
Corrected comparison (native tools path, same 4-tool battery, greedy):
| model | tier-1 fired | tier-1 name=todo | tier-2 todo-first | tier-3 fired (want 0) | tier-4 fired (want 0) |
|---|---|---|---|---|---|
| base (no fine-tune) | 18/20 | 17/20 | 3/10 | 5/10 | 0/10 |
| v28 ckpt-175 (this model) | 17/20 | 10/20 | 8/10 | 9/10 | 4/10 |
| armB2-175 (clean retrain) | 10/20 | 8/20 | 10/10 | 4/10 | 0/10 |
What the fine-tune lineage actually changed: todo-first planning on multi-step tasks (tier-2 name=todo 3/10 base to 8/10 v28 to 10/10 armB2). That is real, and it is the one capability the lineage added. The costs: restraint regressions (tier-3 over-trigger 5/10 to 9/10; tier-4 tool-drift on trivia 0/10 to 4/10) and, in armB2, over-suppression (tier-1 10/20).
Status: the fine-tune program is frozen as of 2026-09-03. The base model is the author's runtime default. This checkpoint stays published as the strongest measured demonstration of SFT-taught todo-first planning in the lineage, and as the artifact that exposed the harness error. The Evaluation section below describes behavior on the synthetic interface; do not read it as native-interface competence.
Reproduction: scripts/native_battery.py + scripts/todo_probes.jsonl
scripts/no_drift_probes.jsonl produce the corrected table (any
GGUF, CTX=65536 to match the serve config).qwen3.8-9b-hermes-fc-todo),
QLoRA adapter on top (envelope + tool tokens already baked in)hermes-fc-v28-data); probes
were kept virgin (zero overlap with generation or training)<|tool_call|>, <|tool_response|>)Qwen3_5ForConditionalGeneration with
text_config.vocab_size = 248079. Visual tower dropped (text-only).This model is intended for experimenters working on tool-call behavior in 9B-class models. It is a research artifact, not a product.
scripts/)Independent evaluation of this checkpoint is not complete. The training
data, scripts, and harness in scripts/ are published so that anyone —
including the author — can evaluate it. Do not treat this model as
benchmarked or validated for any specific behavior.
The training data itself is not published. The rows derive from the author's own agent sessions, and some rows contain private strings (hostnames, session identifiers). The scripts that build the dataset are published; the data is not. The weights do not contain those strings.
All scripts that produced this model are in scripts/:
| file | purpose |
|---|---|
build_v2.py / gapfill_v2.py |
dataset build from mined trajectories (the 979 rows) |
qwen38_train_v2.py |
QLoRA training via Unsloth (246 planned steps) |
merge_qwen38_v2_pure.py |
shard-streamed BF16 merge (pure torch; vision tower dropped; embedding/lm_head resized with EOS-anchor rows) |
subagent-qwen38-workloads.py + subagent-qwen38.sh |
tool-call evaluation harness (workloads A–E) |
The scripts carry the author's machine paths (F:/models/..., C:/AI/...)
and run on this rig's ROCm stack (Unsloth 2026.8.18, torch 2.11 ROCm,
RX 7700 XT); adjust paths for your environment.
Convert/quantize: llama-rocmfpx fork, convert_hf_to_gguf.py --outtype bf16
→ llama-quantize Q4_0_ROCMFP4_FAST (see GGUF repo).
Q4_0_ROCMFP4_FAST GGUF (ROCmFPX format, for AMD RDNA3 kernels) is published
in the companion repo. ROCmFPX quants target AMD ROCm inference; for portable
use, convert from the BF16 merge here.
Base model and license inherited from Empero/Qwen3.8-9B (Apache-2.0). Training data derived from the author's own Hermes agent sessions.
model-00011-of-00012.safetensors 2.0 GBmodel-00012-of-00012.safetensors 2.0 GBmodel-00009-of-00012.safetensors 1.5 GBmodel-00008-of-00012.safetensors 1.5 GBmodel-00006-of-00012.safetensors 1.5 GBmodel-00002-of-00012.safetensors 1.4 GBmodel-00004-of-00012.safetensors 1.4 GBmodel-00005-of-00012.safetensors 1.4 GBmodel-00001-of-00012.safetensors 1.4 GBmodel-00003-of-00012.safetensors 1.4 GBmodel-00007-of-00012.safetensors 1.4 GBmodel-00010-of-00012.safetensors 1.4 GBHave the files? Submit a torrent.
Submit a torrentHave the complete files and permission to share them? Replace the example folder path. The script checks the files, makes a torrent and submits its magnet and checksums. Keep your torrent client seeding. Get your community key from your settings, and read the script before running it.
curl -fsSL https://pirateface.co/package.sh | bash -s -- --repo JamieBradfield/qwen3.8-9b-hermes-fc-real-traces ./model-folder