DistilGpt2Conv1d
A GPT-2 Conv1D projection stored in Hugging Face [input, output] weight layout.
Record fields
| Name | Description |
|---|---|
Bias | |
Weight |
Instance members
| Name | Description |
|---|---|
this.forward input | Apply the affine projection without transposing its stored weight. |
DistilGpt2Attention
Projection layers in one DistilGPT-2 self-attention block.
Record fields
| Name | Description |
|---|---|
Output | |
Qkv |
DistilGpt2Mlp
Projection layers in one DistilGPT-2 feed-forward block.
Record fields
| Name | Description |
|---|---|
Input | |
Output |
DistilGpt2Block
One DistilGPT-2 transformer block.
Record fields
| Name | Description |
|---|---|
Attention | |
Mlp | |
Norm1 | |
Norm2 |
DistilGpt2
Construction, state, cache, and forward operations for DistilGPT-2.
| Function | Description |
|---|---|
asCausalLm model | Bind a DistilGPT-2 model to the common typed causal language-model interface. |
create config dtype device | Create a DistilGPT-2 model from a validated configuration. |
createCache batchSize capacity model | Allocate a reusable fixed-capacity key/value cache for a model. |
descriptor | State descriptor whose canonical names and shapes match Hugging Face DistilGPT-2 weights. |
dispose model | Dispose tensors owned by a DistilGPT-2 model. |
forward input model | Run DistilGPT-2. When a cache is supplied, only new tokens should be passed after prefill. |
loadConfig path | Load and validate a Hugging Face config.json file. |
loadFromDirectory directory device | Load config and a strict single-file or sharded SafeTensors state from a local directory. |
state model | Create a validated named state view using Hugging Face weight names. |
asCausalLm
asCausalLm modelBind a DistilGPT-2 model to the common typed causal language-model interface.
Parameters
model:DistilGpt2
Returns CausalLm<DistilGpt2Cache>
create
create config dtype deviceCreate a DistilGPT-2 model from a validated configuration.
Parameters
config:DistilGpt2Configdtype:ScalarTypedevice:Device
Returns DistilGpt2
createCache
createCache batchSize capacity modelAllocate a reusable fixed-capacity key/value cache for a model.
Parameters
batchSize:int64capacity:int64model:DistilGpt2
Returns DistilGpt2Cache
descriptor
descriptorState descriptor whose canonical names and shapes match Hugging Face DistilGPT-2 weights.
Returns ModelDescriptor<DistilGpt2>
dispose
dispose modelDispose tensors owned by a DistilGPT-2 model.
Parameters
model:DistilGpt2
forward
forward input modelRun DistilGPT-2. When a cache is supplied, only new tokens should be passed after prefill.
Parameters
input:DistilGpt2Inputmodel:DistilGpt2
Returns DistilGpt2Output
loadConfig
loadConfig pathLoad and validate a Hugging Face config.json file.
Parameters
path:string
Returns DistilGpt2Config
loadFromDirectory
loadFromDirectory directory deviceLoad config and a strict single-file or sharded SafeTensors state from a local directory.
Parameters
directory:stringdevice:Device
Returns DistilGpt2 * LoadReport
state
state modelCreate a validated named state view using Hugging Face weight names.
Parameters
model:DistilGpt2
Returns ModelState