GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
k33wu
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
k33wu
Model Tree
Qwen/Qwen3-1.7B
Input Modalities
Output Modalities
Supported Functionality
The adapter was tested with a mask: before each token is picked, every token whose text
contains e, E or a non-ASCII character (so look-alikes such as é can't sneak in) gets a score of
negative infinity, so it can never be chosen. Special tokens such as end-of-answer are kept. The mask
blocks 103,639 of Qwen's 151,936 tokens.
The mask is what guarantees there are no e's. Without it, this adapter avoided e in only 9% of answers.
sh
pip install torch transformers peft acceleratepython
import torchfrom peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer def build_forbidden_token_mask(tokenizer, vocab_size, forbidden_chars='eE', ascii_only=False, device=None, preserve_added_tokens=False): """Mask decoded token text while preserving special tokens such as EOS.""" special_ids = set(tokenizer.all_special_ids) if preserve_added_tokens: special_ids.update(tokenizer.added_tokens_decoder) texts = tokenizer.batch_decode([[i] for i in range(vocab_size)]) return torch.tensor([ i not in special_ids and ( any(c in forbidden_chars for c in text) or (ascii_only and not text.isascii()) ) for i, text in enumerate(texts) ], dtype=torch.bool, device=device) tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen3-1.7B')model = AutoModelForCausalLM.from_pretrained('Qwen/Qwen3-1.7B', dtype='auto', device_map='auto')model = PeftModel.from_pretrained(model, 'k33wu/qwen3-1.7b-no-e-sft180') # the same mask used in testing; bad_words_ids blocks each of these tokensforbidden = build_forbidden_token_mask(tokenizer, model.config.vocab_size, 'eE', ascii_only=True)bad_words_ids = [[i] for i in forbidden.nonzero().flatten().tolist()] messages = [ {'role': 'system', 'content': "Answer in natural English without the letters 'e' or 'E'. " "Preserve correct spelling and answer the user's request. " "This restriction applies only to your final answer, not your thinking."}, {'role': 'user', 'content': 'Describe the ocean.'},]ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, enable_thinking=False, return_tensors='pt', return_dict=True).to(model.device)# temperature 0.7 over all tokens (no top-k/top-p), as in the eval belowout = model.generate(**ids, bad_words_ids=bad_words_ids, max_new_tokens=300, do_sample=True, temperature=0.7, top_k=0, top_p=1.0)print(tokenizer.decode(out[0, ids['input_ids'].shape[1]:], skip_special_tokens=True))Use the system prompt above and keep thinking off (enable_thinking=False). That is how the adapter
was trained and tested.
Example answers (greedy decoding):
Describe the ocean. Ocans hold vast liquid, with a broad floor and a high wall. Many fish and animals swim in it. It has a strong pull, and its floor can shift.
Give me three tips for a job interview.
- Know your job and your company. 2. Talk about what you can do, not just what you know. 3. Ask a good follow-up and thank you for your job.
The eval uses 50 chat prompts that can be answered without e, with 4 samples each (200 answers),
temperature 0.7 and a 300-token cap. An answer counts as usable if it has no e and an LLM judge
(gpt-6-luna) gives it at least 4 out of 5 for both coherence and responding to the prompt.
| Model | Usable, mask on | No e, mask off | Usable, mask off |
|---|---|---|---|
| Qwen3-1.7B, no adapter | 26.0% | 0% | 0% |
| This adapter | 35.0% | 9.0% | 4.5% |
| SFT, then 300 steps of RL (qwen3-1.7b-no-e) | 60.5% | not measured | not measured |
q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj) in all 28 layers.
17.4M parameters, stored in float32. The base model ran in bf16.import torchfrom peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer def build_forbidden_token_mask(tokenizer, vocab_size, forbidden_chars='eE', ascii_only=False, device=None, preserve_added_tokens=False): """Mask decoded token text while preserving special tokens such as EOS.""" special_ids = set(tokenizer.all_special_ids) if preserve_added_tokens: special_ids.update(tokenizer.added_tokens_decoder) texts = tokenizer.batch_decode([[i] for i in range(vocab_size)]) return torch.tensor([ i not in special_ids and ( any(c in forbidden_chars for c in text) or (ascii_only and not text.isascii()) ) for i, text in enumerate(texts) ], dtype=torch.bool, device=device) tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen3-1.7B')model = AutoModelForCausalLM.from_pretrained('Qwen/Qwen3-1.7B', dtype='auto', device_map='auto')model = PeftModel.from_pretrained(model, 'k33wu/qwen3-1.7b-no-e-sft180') # the same mask used in testing; bad_words_ids blocks each of these tokensforbidden = build_forbidden_token_mask(tokenizer, model.config.vocab_size, 'eE', ascii_only=True)bad_words_ids = [[i] for i in forbidden.nonzero().flatten().tolist()] messages = [ {'role': 'system', 'content': "Answer in natural English without the letters 'e' or 'E'. " "Preserve correct spelling and answer the user's request. " "This restriction applies only to your final answer, not your thinking."}, {'role': 'user', 'content': 'Describe the ocean.'},]ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, enable_thinking=False, return_tensors='pt', return_dict=True).to(model.device)# temperature 0.7 over all tokens (no top-k/top-p), as in the eval belowout = model.generate(**ids, bad_words_ids=bad_words_ids, max_new_tokens=300, do_sample=True, temperature=0.7, top_k=0, top_p=1.0)print(tokenizer.decode(out[0, ids['input_ids'].shape[1]:], skip_special_tokens=True))