GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
ldp72
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
ldp72
Model Tree
Base
Orange/Qwen3-8B
Fine-tuned
this model
Input Modalities
Text
Output Modalities
Text
Supported Functionality
Dedicated Endpoints
Container
This model can be used with the transformers library using pipeline abstraction as follows:
python
import torchfrom transformers import pipelinemodel_id = "Orange/Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1"pipe = pipeline("text-generation",model=model_id,torch_dtype=torch.bfloat16,device_map="auto",)messages = [{"role": "system", "content": "You are chatbot specialized on Wolof language."},{"role": "user", "content": "Can you give a sample of your specialized knowledge?"},]outputs = pipe(messages,max_new_tokens=256,)print(outputs[0]["generated_text"][-1])
[More Information Needed]
[More Information Needed]
[More Information Needed]
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
Use the code below to get started with the model.
[More Information Needed]
This model was finetuned with Orange internal fine tuning tools with the Docker Image tagged 0.2.6t5alpha in the registry and the following configuration file:
yaml
data:add_dataset_name: falsedataset_name:train:- path: wolof-lm/en_function_calling-instructionsrevision: v10split: train- path: wolof-lm/alpaca_translated_into_wolof-instructionsrevision: 2025.02.28split: train- path: wolof-lm/argilla_databricks-dolly-15k-curated-multilingual_en_translated_into_wolof-instructionsrevision: 2025.02.28split: train- path: wolof-lm/samsung_samsum_instruct_dialogues_translated_in_wolof-instructionsrevision: 2025.02.28split: train- path: wolof-lm/alwaly_french_wolof-instructionssplit: train- path: wolof-lm/alwaly_wolof_to_french-instructionssplit: train- path: wolof-lm/claire_masking_multilogue-instructionssplit: train- path: wolof-lm/huggingfaceh4_helpful-anthropic-raw_train_translated_in_wolof-instructionssplit: train- path: wolof-lm/lodrick-the-lafted-hermes-217k-instructionssplit: train- path: wolof-lm/wisenut-nlp-team_translated_in_wolof-instructionssplit: train- path: wolof-lm/claire_one_to_one_dialogue-instructionsrevision: 2026.01.16split: train- path: wolof-lm/wol-sonatel-2025-conversationsrevision: 2026.02.18split: trainvalidation_conversational_eng:- path: wolof-lm/oasst1-en-instructionsrevision: 2026.03.19split: validationvalidation_conversational_fra:- path: wolof-lm/oasst1-fr-instructionsrevision: 2026.03.19split: validation- path: wolof-lm/synthetic-fr-instructionsrevision: 2026.03.19split: validationvalidation_conversational_wol:- path: wolof-lm/claire_one_to_one_dialogue-instructionsrevision: 2026.01.16split: validation- path: wolof-lm/wol-sonatel-2025-conversationsrevision: 2026.02.18split: validationvalidation_fc:- path: wolof-lm/en_function_calling-instructionsrevision: v10split: validationvalidation_if:- path: wolof-lm/tulu-3-sft-personas-instruction-followingsplit: validationvalidation_it-v2:- path: wolof-lm/alpaca_translated_into_wolof-instructionsrevision: 2025.02.28split: validation- path: wolof-lm/argilla_databricks-dolly-15k-curated-multilingual_en_translated_into_wolof-instructionsrevision: 2026.03.02split: validation- path: wolof-lm/samsung_samsum_instruct_dialogues_translated_in_wolof-instructionsrevision: 2025.02.28split: validation- path: wolof-lm/alwaly_french_wolof-instructionssplit: validation- path: wolof-lm/alwaly_wolof_to_french-instructionssplit: validation- path: wolof-lm/claire_masking_multilogue-instructionssplit: validation- path: wolof-lm/huggingfaceh4_helpful-anthropic-raw_train_translated_in_wolof-instructionssplit: validation- path: wolof-lm/lodrick-the-lafted-hermes-217k-instructionssplit: validation- path: wolof-lm/wisenut-nlp-team_translated_in_wolof-instructionssplit: validationdebug: falseimplementation_name: conversationsdescription:contributors:- email: claire.perroux@orange.comfirst_name: Clairelast_name: Perroux- email: pierre.adam@orange.comfirst_name: Pierrelast_name: Adamdomain: Woloflanguages:- womodel_name: Orange/Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1image:name: registry.gitlab.tech.orange/nepal/knowledge/orangelm/lm-adaptation/lm-adaptationversion: 0.2.6t5alphamodel:attn_implementation: flash_attention_2chat_template_tokenizer: Orange/Qwen3-8Bmodel_name_or_path: Orange/Qwen3-8Btrust_remote_code: truesoftware_info:git:branch: 119-cuda-13-and-transformers-5commit_date: '2026-06-10 15:41:20+02:00'commit_hash: 314a8efb580bf36117958e80dc33d7e221aa7d30commit_message: 'build: set same version for kernels library in pyprojectas in requirements.txt'is_dirty: falseshort_hash: 314a8efparameters:accelerate_file: nullclean_output: falsecluster: marcelcodecarbon_log_level: warningcollator_stats: falseconfig_file: resources/configs/Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1.ymlconstraints: queue:name=gpu_mem_80,gpu:2,partition:name=multigpu,partition:name=Alldata_seed: 42debug: falsedeepspeed_file: resources/deepspeed/zero3.jsondry_run: falseenvironment_variable:TOKENIZERS_PARALLELISM: 'false'load_only: nulllog_level: INFOlog_output: nullmerge_peft: falseno_requeue: falseoutput_dir: Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1-t5-torch28-2gpuprofiler_config: nullpush_to_hub: falseqat_config: nullrequeue_n_epochs: 1time_out: 72htime_requeue: falseuse_qat: falseversion: 0.21.2.dev33+gb8ccade7f.d20260610training:assistant_only_loss: truebf16: truedataloader_num_workers: 4dataloader_persistent_workers: falsedataloader_pin_memory: falsedataloader_prefetch_factor: 2deepspeed: /config/zero3.jsondisable_tqdm: trueeval_accumulation_steps: 1eval_steps: 10eval_strategy: stepsfp16: falsegradient_accumulation_steps: 32gradient_checkpointing: truelearning_rate: 2.0e-05log_level: warninglogging_dir: /outputs/Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1-t5-torch28-2gpu/logslogging_steps: 1lr_scheduler_type: cosinemax_grad_norm: 1.0max_length: 2048max_steps: -1num_train_epochs: 3optim: paged_adamw_32bitoutput_dir: /outputs/Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1-t5-torch28-2gpuper_device_eval_batch_size: 16per_device_train_batch_size: 16push_to_hub: falsereport_to: tensorboardsave_steps: 0save_strategy: epochsave_total_limit: 1seed: 42torch_compile: falsetraining_type: instruct-tuninguse_liger_kernel: truewarmup_ratio: 0.05weight_decay: 0.1
The model was trained on 2 gpus with at least 80GB on each gpu.
The model was trained using deepspeed with the following configuration file:
json
{"fp16": {"enabled": "auto","loss_scale": 0,"loss_scale_window": 1000,"initial_scale_power": 16,"hysteresis": 2,"min_loss_scale": 1},"bf16": {"enabled": "auto"},"zero_optimization": {"stage": 3,"offload_optimizer": {"device": "cpu","pin_memory": false},"overlap_comm": true,"contiguous_gradients": true,"sub_group_size": "1e9","reduce_bucket_size": "auto","stage3_prefetch_bucket_size": "auto","stage3_param_persistence_threshold": "auto","stage3_max_live_parameters": "1e9","stage3_max_reuse_distance": "1e9","stage3_gather_16bit_weights_on_model_save": true},"gradient_accumulation_steps": "auto","gradient_clipping": "auto","steps_per_print": 2000,"train_batch_size": "auto","train_micro_batch_size_per_gpu": "auto","wall_clock_breakdown": false}
This model was trained on the following datasets:
yaml
- path: wolof-lm/en_function_calling-instructionsrevision: v10split: train- path: wolof-lm/alpaca_translated_into_wolof-instructionsrevision: 2025.02.28split: train- path: wolof-lm/argilla_databricks-dolly-15k-curated-multilingual_en_translated_into_wolof-instructionsrevision: 2025.02.28split: train- path: wolof-lm/samsung_samsum_instruct_dialogues_translated_in_wolof-instructionsrevision: 2025.02.28split: train- path: wolof-lm/alwaly_french_wolof-instructionssplit: train- path: wolof-lm/alwaly_wolof_to_french-instructionssplit: train- path: wolof-lm/claire_masking_multilogue-instructionssplit: train- path: wolof-lm/huggingfaceh4_helpful-anthropic-raw_train_translated_in_wolof-instructionssplit: train- path: wolof-lm/lodrick-the-lafted-hermes-217k-instructionssplit: train- path: wolof-lm/wisenut-nlp-team_translated_in_wolof-instructionssplit: train- path: wolof-lm/claire_one_to_one_dialogue-instructionsrevision: 2026.01.16split: train- path: wolof-lm/wol-sonatel-2025-conversationsrevision: 2026.02.18split: train
[More Information Needed]
SFTTrainer,other parameters were set as default:yaml
assistant_only_loss: truebf16: truedataloader_num_workers: 4dataloader_persistent_workers: falsedataloader_pin_memory: falsedataloader_prefetch_factor: 2deepspeed: /config/zero3.jsondisable_tqdm: trueeval_accumulation_steps: 1eval_steps: 10eval_strategy: stepsfp16: falsegradient_accumulation_steps: 32gradient_checkpointing: truelearning_rate: 2.0e-05log_level: warninglogging_dir: /outputs/Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1-t5-torch28-2gpu/logslogging_steps: 1lr_scheduler_type: cosinemax_grad_norm: 1.0max_length: 2048max_steps: -1num_train_epochs: 3optim: paged_adamw_32bitoutput_dir: /outputs/Wolof-Qwen3-8B-it-v2-fc-v2-conv-v1-t5-torch28-2gpuper_device_eval_batch_size: 16per_device_train_batch_size: 16push_to_hub: falsereport_to: tensorboardsave_steps: 0save_strategy: epochsave_total_limit: 1seed: 42torch_compile: falseuse_liger_kernel: truewarmup_ratio: 0.05weight_decay: 0.1
Number of Tokens vs Steps
Learning Rate Curve
Training Loss
Validation Loss
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
emissions.csv (emissions were computed using codecarbon)[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
BibTeX:
[More Information Needed]
APA:
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
Thanks to Claire Perroux, Pierre Adam for adding this model.