GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
saidutta69
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
saidutta69
Model Tree
utter-project/EuroLLM-1.7B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Trial 75 of a 200-trial Heretic 2.0 ARA run (seed 42).
| Parameter | Value |
|---|---|
| start_layer_index | 8 |
| end_layer_index | 23 |
| preserve_good_behavior_weight | 0.0554 |
| steer_bad_behavior_weight | 0.0001 |
| overcorrect_relative_weight | 0.5514 |
| neighbor_count | 1 |
| Metric | This model | Original model (utter-project/EuroLLM-1.7B-Instruct) |
|---|---|---|
| Refusals | 3/100 | 3/100 |
| KL divergence | 0.0113 | 0 (by definition) |
All three remaining refusals stayed exactly as they were: the base was already at 3/100, so there was nothing to remove. KL 0.0113 is a small weight change, and the lm-eval run below is unchanged within noise. Do not read this as a "high-fidelity edit that also decensored the model" - it only did the first half.
EuroLLM's published numbers for the 1.7B base, not re-run against this checkpoint:
| Benchmark | EuroLLM-1.7B-Instruct |
|---|---|
| FLORES-200 (AVG, 71 pairs) | 86.89 |
| WMT-23 (AVG) | 82.91 |
| WMT-24 (AVG) | 79.32 |
| Arc Challenge (AVG, 17 languages) | 0.3496 |
| HellaSwag (AVG, 16 languages) | 0.4744 |
Abliteration removes a weight subspace; it does not add capability. Treat these as the base model's numbers, not a re-measurement of this checkpoint.
User-run lm-eval 0.4.13, this checkpoint vs the base (MMLU: 14,042 questions, stderr 0.37 pp; IFEval prompt-level stderr 1.65 pp):
| Benchmark | Metric | This model | Base |
|---|---|---|---|
| IFEval | prompt-level strict acc | 0.1793 | 0.1756 |
| IFEval | prompt-level loose acc | 0.1830 | 0.1830 |
| IFEval | inst-level strict acc | 0.3165 | 0.3165 |
| IFEval | inst-level loose acc | 0.3201 | 0.3249 |
| MMLU | acc | 0.2592 | 0.2576 |
Every delta is inside one standard error. This is the expected result for a 3->3/100 refusal run.
Directional ablation finds one refusal direction and orthogonalizes attention/MLP outputs against
it at a fixed strength on every layer. ARA learns a low-rank subspace over a layer range and blends
a "preserve good behaviour" objective on mlabonne/harmless_alpaca with a "steer bad behaviour"
objective on mlabonne/harmful_behaviors. This run weighted the preserve objective 0.0554 against
a 0.0001 steering weight, and the result is a small edit (KL 0.0113) that leaves refusal behaviour
where it started. See the Heretic repo.
| File | Size |
|---|---|
model.safetensors | 3.31 GB |
BF16, 1,656,850,432 parameters (1.66B), untied input/output embeddings (~0.262B each). The
reproduce/ directory carries the full Heretic 2.0 recipe - config.toml, requirements.txt, the
Optuna study journal, and SHA-256 sums - so this exact checkpoint can be regenerated bit-for-bit
with heretic --reproduce reproduce/reproduce.json.
Test run, so F16 is the only GGUF published (3.32 GB). It was produced with
llama.cpp master on 2026-10-30; no quant ladder was
generated for this checkpoint. Quantize the F16 yourself with llama-quantize if you need smaller
files.
| File | Format | Size |
|---|---|---|
EuroLLM-1.7B-Instruct-heretic-F16.gguf | GGUF F16 | 3.32 GB |
Text GGUF: arch llama, 219 tensors, 24 blocks, 4096 context,
128,000 vocab, untied embeddings, RoPE theta 10,000. The tensor count and
1,656,850,432 elements match the safetensors exactly, so nothing was dropped. Standard dense Llama
architecture - loads natively in llama.cpp, Ollama, LM Studio and Jan, and works with MLX and vLLM.
Run llama serve -hf saidutta69/EuroLLM-1.7B-Instruct-heretic:F16 to pull the F16 file explicitly - there is no smaller quant in
this repo, so the bare -hf <repo> form has nothing else to pick.
bash
# llama.cpp - ChatML template needs --jinjallama serve -hf saidutta69/EuroLLM-1.7B-Instruct-heretic:F16 --jinja -c 4096python
# transformersfrom transformers import AutoModelForCausalLM, AutoTokenizer model_id = "saidutta69/EuroLLM-1.7B-Instruct-heretic"tok = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained(model_id, dtype="auto", device_map="auto") messages = [ {"role": "system", "content": "You are EuroLLM, an AI assistant specialised in European languages."}, {"role": "user", "content": "Translate to Polish: 'The library closes at six.'"},]inputs = tok.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device)out = model.generate(**inputs, max_new_tokens=256)print(tok.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))Abliteration does not touch the tokenizer or the chat template, so the base model's multilingual setup is intact.
generation_config.json sets do_sample: true with no temperature, top-p or top-k override, and
the base card publishes no tuned sampling settings, so generate() runs on the model defaults.
For translation, a low temperature is the safe starting point; validate on your own language pairs
rather than copying a recipe from a different model.
ChatML: system, user and assistant turns are wrapped in <|im_start|>role ... <|im_end|>, and the
template is shipped as chat_template.jinja. The base card's example uses an empty system turn; a
language hint there is a reasonable way to steer output. There is no thinking mode, no
enable_thinking flag, and no tool-call schema - this is the plain EuroLLM instruct interface.
| Device | File | Weights |
|---|---|---|
| Laptop / desktop (8 GB+ RAM) | F16 | 3.32 GB |
| 8-12 GB GPU | F16 (full offload) | 3.32 GB |
| CPU-only server | F16 | 3.32 GB |
F16 only, as published here. At 1.66B and 4K context the KV cache is small, so decode is bandwidth-bound and a quantized build (which you would have to make yourself) runs acceptably on a phone-class CPU.
| Architecture | LlamaForCausalLM (decoder-only dense transformer) |
| Parameters | 1,656,850,432 (1.66B) |
| Layers / heads | 24 layers, 16 attention heads, 8 KV heads (GQA), head_dim 128 |
| Hidden / intermediate | 2,048 / 5,632 (SwiGLU, pre-layer RMSNorm) |
| Position embedding | RoPE, theta = 10,000 |
| Context length | 4,096 native - no rope_scaling in the config |
| Vocab | 128,000 |
| Precision | bfloat16, untied embeddings |
| Languages | 35: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Irish, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish, Arabic, Catalan, Chinese, Galician, Hindi, Japanese, Korean, Norwegian, Russian, Turkish, Ukrainian |
| Developers | Unbabel, Instituto Superior Tecnico, Instituto de Telecomunicacoes, University of Edinburgh, Aveni, University of Paris-Saclay, University of Amsterdam, Naver Labs, Sorbonne Universite - funded by the European Union |
| Training | 4 trillion tokens across the language set, including parallel en-xx and xx-en data; 256 NVIDIA H100s on Marenostrum 5; instruction-tuned on EuroBlocks |
| Base model | utter-project/EuroLLM-1.7B-Instruct (a post-trained utter-project/EuroLLM-1.7B) |
This is a weight edit, not an alignment layer. Refusal behaviour is unchanged from the base (3/100), so it does not add or remove safety behaviour, and the base is documented as "not aligned to human preferences". It inherits EuroLLM's factual limitations and biases across all 35 languages; abliteration removes a direction in weight space, it does not add capability or judgment. Validate translation output before shipping it, especially on low-resource pairs and on untrusted content that you would not want laundered past another system's filters.
Made with ❤️ by RACER IS OP - follow for more uncensored models
Inherits the apache-2.0 license from the base
model, which is tagged Apache-2.0 on the Hub but ships no LICENSE file, so the canonical
license text is linked rather than a repo blob. This is a derivative work; keep the license and
attribution when redistributing, and do not imply endorsement by the EuroLLM authors.
Recorded in reproduce/reproduce.json:
Linux-4.19.0-gvisor-x86_64-with-glibc2.41Heretic notes that KL-divergence and refusal numbers can vary slightly with hardware and library versions.
# llama.cpp - ChatML template needs --jinjallama serve -hf saidutta69/EuroLLM-1.7B-Instruct-heretic:F16 --jinja -c 4096# transformersfrom transformers import AutoModelForCausalLM, AutoTokenizer model_id = "saidutta69/EuroLLM-1.7B-Instruct-heretic"tok = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained(model_id, dtype="auto", device_map="auto") messages = [ {"role": "system", "content": "You are EuroLLM, an AI assistant specialised in European languages."}, {"role": "user", "content": "Translate to Polish: 'The library closes at six.'"},]inputs = tok.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device)out = model.generate(**inputs, max_new_tokens=256)print(tok.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))