Triangle104/TQ2.5-14B-Neon-v1-Q8_0-GGUF

This model was converted to GGUF format from allura-org/TQ2.5-14B-Neon-v1 using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model.

---

Model details:

RP finetune of Supernova-Medius. Turned out surprisingly nice on it's own, I honestly made it only as a merge fuel, but it impressed me and Prodeus enough to release it separately (history repeats I guess, Sugarquill also started out this way). Quite interesting prose, definitely quite distinct from Supernova or EVA for that matter. Instruction following is decent as well. Not really much to say about this one, just a decent RP model, tbh. Euryale-inspired I guess.

Model was trained by Auri.

Training notes

Model was trained on a dataset consisting of 77M tokens of synthetic RP and short story gen data. Training took around 2 hours on 8xH100 SXM node. Training config was more or less reused from Sugarquill, and it worked fairly well again. Had the node crash after finishing the training and merging in the LoRA, so I had to merge it with MergeKit on a separate node, otherwise everything was smooth.

Huge thanks to Retis Labs for sponsoring this run!

Format

Model responds to ChatML instruct formatting, exactly like it's base model.

<|im_start|>system {system message}<|im_end|> <|im_start|>user {user message}<|im_end|> <|im_start|>assistant {response}<|im_end|>

Recommended Samplers

My classic stable Qwen setup works quite well:

Temperature - 0.8 Min-P - 0.05 Top-A - 0.3 Repetition Penalty - 1.03

Training config See Axolotl config

axolotl version 0.6.0

Model

base_model: arcee-ai/SuperNova-Medius strict: false

Liger Kernels (optimization)

plugins:

• axolotl.integrations.liger.LigerPlugin liger_rope: true liger_rms_norm: true liger_swiglu: true liger_fused_linear_cross_entropy: true

Output and HuggingFace

output_dir: /workspace/axolotl/TQ-2.5-14B-Neon hub_model_id: allura-org/TQ-2.5-14B-Neon-LoRA hf_use_auth_token: true hub_strategy: "all_checkpoints"

WandB

wandb_project: allura-org wandb_entity: wandb_name: TQ-2.5-14B-Neon-1

Data

chat_template: chatml #train_on_inputs: false group_by_length: false datasets:

• path: allura-org/neon-41k type: chat_template field_messages: conversations message_field_role: from message_field_content: value

Evaluation

val_set_size: 0.01 evals_per_epoch: 4 eval_table_size: eval_max_new_tokens: 128

Technical aspects

sequence_len: 16384 save_safetensors: true saves_per_epoch: 2 logging_steps: 1 special_tokens:

Quantization

bf16: auto fp16: tf32: false

For LoRA

load_in_8bit: false load_in_4bit: false

LoRA

peft_use_rslora: true peft_use_dora: false # better but slower adapter: lora # lora or qlora lora_model_dir: lora_r: 64 # 64 is optimal for most trains on instruct lora_alpha: 32 lora_dropout: 0.1 lora_target_linear: true lora_fan_in_fan_out: lora_target_modules:

- embed_tokens

- lm_head

#loraplus_lr_ratio: 8 # works to converge faster but is kinda cancer bc makes model unstable #loraplus_lr_embedding:

Training hyperparameters

max_steps:

num_epochs: 2

Anti Overfit and Stability

weight_decay: 0.01 max_grad_norm: 1.0

Learning Rate

warmup_ratio: 0.05 learning_rate: 0.00003 lr_scheduler: cosine #lr_scheduler_kwargs:

min_lr: 0.0000024

optimizer: paged_ademamix_8bit # usually adamw_torch or paged_adamw_8bit

Batch Size

gradient_accumulation_steps: 4 # More effective batch size - stabler train, usually. MBS also speeds it up. micro_batch_size: 4 # Batch size per gpu = micro_batch_size * gradient_accumulation_steps eval_batch_size: 1

Optimizations

pad_to_sequence_len: true sample_packing: true eval_sample_packing: false flash_attention: true xformers_attention: gradient_checkpointing: "unsloth" gradient_checkpointing_kwargs: use_reentrant: true local_rank: deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json # Only use with multi gpu # _bf16_cpuoffload_all

fsdp:

- full_shard

- auto_wrap

fsdp_config:

fsdp_limit_all_gathers: true

fsdp_sync_module_states: true

fsdp_offload_params: true

fsdp_use_orig_params: false

fsdp_cpu_ram_efficient_loading: true

fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP

fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer

fsdp_state_dict_type: FULL_STATE_DICT

fsdp_sharding_strategy: FULL_SHARD

Misc

early_stopping_patience: debug:

---

Use with llama.cpp

Install llama.cpp through brew (works on Mac and Linux)

brew install llama.cpp

Invoke the llama.cpp server or the CLI.

CLI:

llama-cli --hf-repo Triangle104/TQ2.5-14B-Neon-v1-Q8_0-GGUF --hf-file tq2.5-14b-neon-v1-q8_0.gguf -p "The meaning to life and the universe is"

Server:

llama-server --hf-repo Triangle104/TQ2.5-14B-Neon-v1-Q8_0-GGUF --hf-file tq2.5-14b-neon-v1-q8_0.gguf -c 2048

Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well.

Step 1: Clone llama.cpp from GitHub.

git clone https://github.com/ggerganov/llama.cpp

Step 2: Move into the llama.cpp folder and build it with LLAMA_CURL=1 flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).

cd llama.cpp && LLAMA_CURL=1 make

Step 3: Run inference through the main binary.

./llama-cli --hf-repo Triangle104/TQ2.5-14B-Neon-v1-Q8_0-GGUF --hf-file tq2.5-14b-neon-v1-q8_0.gguf -p "The meaning to life and the universe is"

or

./llama-server --hf-repo Triangle104/TQ2.5-14B-Neon-v1-Q8_0-GGUF --hf-file tq2.5-14b-neon-v1-q8_0.gguf -c 2048