yueqis

This model is a fine-tuned version of yueqis/web-qwen-coder-14b-3epochs-25k-5e-5 on the web dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 64 - optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1.0 - Transformers 4.57.1 - Pytorch 2.8.0+cu128 - Datasets 4.0.0 - Tokenizers 0.22.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-32B-Instruct on the sweonly dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-32B-Instruct on the nonweb dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the nonwebsweagent dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-14B-Instruct on the nonweb dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-14B-Instruct on the nonwebsweagent dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the nonwebmcp dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 64 - optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1.0 - Transformers 4.57.1 - Pytorch 2.8.0+cu128 - Datasets 4.0.0 - Tokenizers 0.22.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the sweonlysweagent dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the sweonlyoriginal5k dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the nonweb dataset. The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 32 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 512 - totalevalbatchsize: 256 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3.0 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the fullsftnonweb dataset. It achieves the following results on the evaluation set: - Loss: 0.3317 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the fullsftv0.4 dataset. It achieves the following results on the evaluation set: - Loss: 0.3446 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 4 - totaltrainbatchsize: 32 - totalevalbatchsize: 8 - optimizer: Use adamwtorchfused with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 | Training Loss | Epoch | Step | Validation Loss | |:-------------:|:------:|:----:|:---------------:| | 0.3427 | 0.4049 | 1000 | 0.3676 | | 0.3254 | 0.8097 | 2000 | 0.3476 | - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the go-browse-wa dataset. It achieves the following results on the evaluation set: - Loss: 0.3853 The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 2 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the go-browse dataset. It achieves the following results on the evaluation set: - Loss: 0.1270 The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 2 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the v0.4 dataset. It achieves the following results on the evaluation set: - Loss: 0.3654 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-14B on the fullsftv0.4 dataset. It achieves the following results on the evaluation set: - Loss: 0.2949 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 4 - totaltrainbatchsize: 32 - totalevalbatchsize: 8 - optimizer: Use adamwtorchfused with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 | Training Loss | Epoch | Step | Validation Loss | |:-------------:|:------:|:----:|:---------------:| | 0.3214 | 0.4049 | 1000 | 0.3258 | | 0.3208 | 0.8097 | 2000 | 0.2988 | - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the nonwebsftv0.4 dataset. It achieves the following results on the evaluation set: - Loss: 0.3102 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the sweonlysweagent dataset. It achieves the following results on the evaluation set: - Loss: 0.2691 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the sweonlymcp dataset. It achieves the following results on the evaluation set: - Loss: 0.2769 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the swe-mix-sample dataset. It achieves the following results on the evaluation set: - Loss: 0.1673 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the fullsftsweagent dataset. It achieves the following results on the evaluation set: - Loss: 0.2499 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the sweonlysweagent dataset. It achieves the following results on the evaluation set: - Loss: 0.2917 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the web-sample dataset. It achieves the following results on the evaluation set: - Loss: 0.4692 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the unifiedagent30K dataset. The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 16 - gradientaccumulationsteps: 8 - totaltrainbatchsize: 128 - totalevalbatchsize: 128 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1.0 - Transformers 4.51.1 - Pytorch 2.6.0+cu124 - Datasets 2.21.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the SWE-smith5kTrajectories dataset. It achieves the following results on the evaluation set: - Loss: 0.1589 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-14B on the unifiedagent30K dataset. The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 64 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1.0 - Transformers 4.51.1 - Pytorch 2.6.0+cu124 - Datasets 2.21.0 - Tokenizers 0.21.1

Qwen3-14B-30K-sft-nopacking-lr1e-5-single-node-val-split This model is a fine-tuned version of Qwen/Qwen3-14B on the unifiedagent30K dataset. It achieves the following results on the evaluation set: - Loss: 0.3302 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.1 - Pytorch 2.6.0+cu124 - Datasets 2.21.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the v0.2 dataset. It achieves the following results on the evaluation set: - Loss: 0.3129 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the sweoriginal dataset. It achieves the following results on the evaluation set: - Loss: 0.0027 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the fullsftnonweb dataset. It achieves the following results on the evaluation set: - Loss: 0.2291 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the fullsftnonweb dataset. It achieves the following results on the evaluation set: - Loss: 0.2784 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the fullsftv0.4 dataset. It achieves the following results on the evaluation set: - Loss: 0.3212 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the fullsftmcp dataset. It achieves the following results on the evaluation set: - Loss: 0.4245 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the fullassistantowl dataset. It achieves the following results on the evaluation set: - Loss: 0.0646 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the fullsftnonweb dataset. It achieves the following results on the evaluation set: - Loss: 0.2527 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the sample-80k dataset. It achieves the following results on the evaluation set: - Loss: 0.3979 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-14B on the unifiedagent30K dataset. The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 16 - gradientaccumulationsteps: 4 - totaltrainbatchsize: 64 - totalevalbatchsize: 128 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1.0 - Transformers 4.51.1 - Pytorch 2.6.0+cu124 - Datasets 3.6.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the unifiedagent30K dataset. The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 16 - gradientaccumulationsteps: 2 - totaltrainbatchsize: 32 - totalevalbatchsize: 128 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1.0 - Transformers 4.51.1 - Pytorch 2.6.0+cu124 - Datasets 3.6.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the code-mix-sample dataset. It achieves the following results on the evaluation set: - Loss: 0.3868 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the SWE-smith5kTrajectories dataset. It achieves the following results on the evaluation set: - Loss: 0.2062 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the go-browse-wa dataset. It achieves the following results on the evaluation set: - Loss: 0.4563 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the unifiedagentdata30kfilteredpolished dataset. It achieves the following results on the evaluation set: - Loss: 0.4546 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the nnetnav-live and the nnetnav-wa datasets. It achieves the following results on the evaluation set: - Loss: 0.9091 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 2 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the fulluserowl dataset. It achieves the following results on the evaluation set: - Loss: 0.3807 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the nonwebmcp dataset. It achieves the following results on the evaluation set: - Loss: 0.2480 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the web-sample-filtered dataset. It achieves the following results on the evaluation set: - Loss: 0.5447 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the sweonlysweagent dataset. It achieves the following results on the evaluation set: - Loss: 0.0843 The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 3 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-14B on the unifiedagent30K dataset. The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 8 - seed: 42 - distributedtype: multi-GPU - numdevices: 16 - gradientaccumulationsteps: 8 - totaltrainbatchsize: 128 - totalevalbatchsize: 128 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1.0 - Transformers 4.51.1 - Pytorch 2.6.0+cu124 - Datasets 2.21.0 - Tokenizers 0.21.1

This model is a fine-tuned version of Qwen/Qwen3-8B on the agenttuning dataset. It achieves the following results on the evaluation set: - Loss: 0.3331 The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 1 - seed: 42 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 128 - totalevalbatchsize: 8 - optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments - lrschedulertype: cosine - lrschedulerwarmupratio: 0.05 - numepochs: 1 - Transformers 4.51.3 - Pytorch 2.7.0+cu126 - Datasets 3.5.0 - Tokenizers 0.21.1