merge
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the passthrough merge method using huihui-ai/Llama-3.2-1B-Instruct-abliterated as a base.
Models Merged
The following models were included in the merge:
Configuration
The following YAML configuration was used to produce this model:
slices:
- sources:
- model: lilmeaty/testing_semifinal
layer_range: [1, 1]
parameters:
weight: 0.3
density: 0.2
gamma: 0.005
normalize: true
int8_mask: true
random_seed: 42
temperature: 0.5
top_p: 0.65
inference: true
max_tokens: 300
stream: true
quantization:
- method: int8
value: 60
- method: int4
value: 40
merge_method: passthrough
base_model: huihui-ai/Llama-3.2-1B-Instruct-abliterated
dtype: float16
compression:
pruning:
enabled: true
sparsity: 0.95
distillation:
enabled: true
temperature: 0.7
model_type: "distilled"
quantization:
enabled: true
methods:
- int8
- int4
inference_optimizations:
caching:
enabled: true
cache_size: 1000
batching:
enabled: true
batch_size: 8
parallelism:
enabled: true
workers: 4
asynchronous:
enabled: true
max_concurrent_tasks: 5
tensor_cores:
enabled: true
gpu:
enabled: true
device: cuda
model_sharding:
enabled: true
shards: 2
memory_optimization:
enabled: true
strategy: "offload"
tensor_compression:
enabled: true
method: "tensor_factorization"
mixture_of_experts:
enabled: true
num_experts: 4
gating_strategy: top_k
top_k: 2
load_balancing:
enabled: true
balance_factor: 0.5
expert_capacity:
max_tokens_per_expert: 512
dynamic_routing:
enabled: true
routing_threshold: 0.1
routing_optimizations:
enabled: true
cache_routing: true
model_sparsity:
enabled: true
sparsity_pattern: "block"
mask_method: "random"
pruning_factor: 0.98
auto_tuning:
enabled: true
batch_size_adaptation:
enabled: true
factor: 0.8
max_batch_size: 32
temperature_scheduling:
enabled: true
start_temp: 1.0
end_temp: 0.5
schedule: "linear"
- Downloads last month
- 26
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.