Laguna-XS-2.1-DFlash
- Engine
- vLLM
- Listed
- 26 Oct 2026
DFlash speculator for poolside/Laguna-XS-2.1.
- Serves
- Chat, Reasoning
- Parameters
- 726M
- Context
- 262,144 tokens
- Weights
- 924 MB
- License
- openmdw-1.1
Capabilities
Read from the model's capability tags. Requests are admitted per tag.
- Chat
chat -
POST /v1/chat/completionsPOST /v1/responses
Endpoint task mode
generate
- Inputs
- Text
- Reasoning effort
reasoning_effort - none, low, medium, high
- Values a request may send; the model family sets the vocabulary.
- Structured output
structured_output - not yet probed
- Admitted until the capability probe observes the model.
- Forced tool choice
forced_tool_choice - not yet probed
- Admitted until the capability probe observes the model.
Launch settings
Every engine serving the model starts with these.
Set by the owner. auto: the serving node derives it. default: the platform's value, left in place.
- Engine
- vLLM
- Context window
--max-model-len - 262,144 tokens
- Quantization
--quantization - none default
- Weights load as stored, bfloat16.
- KV cache dtype
--kv-cache-dtype - auto
- No flag: the checkpoint's KV quantization, else the model dtype.
- Scheduling policy
--scheduling-policy - priority default
- Interactive requests ahead of batch work; same-class traffic in arrival order.
- Tool-call parser
--tool-call-parser - auto
- Resolves to poolside_v1 from the architecture.
- Reasoning parser
--reasoning-parser - auto
- Resolves to poolside_v1 from the architecture.
- Trust remote code
--trust-remote-code - on default
- Python shipped in the bundle may run at load (the default).
Engine tuning
Every setting auto: the serving node derives each one.
Set by the owner. auto: the serving node derives it.
Memory and batching
- GPU memory utilization
--gpu-memory-utilization - auto
- Max sequences
--max-num-seqs - auto
- Max batched tokens
--max-num-batched-tokens - auto
- Long-prefill threshold
--long-prefill-token-threshold - auto
Execution
- Enforce eager
--enforce-eager - auto
- Prefix caching
--enable-prefix-caching - auto
- Async scheduling
--no-async-scheduling - auto
- Custom all-reduce
--disable-custom-all-reduce - auto
Speculative decoding
Draft tokens checked by the model in one pass.
- Speculative decoding
- not configured
- An endpoint's own setting, or the serving node's default for the model family, applies.
Architecture
Read from the model's files at upload.
- Architecture
- laguna
- Parameters
- 725,635,072
- Layers
num_hidden_layers - 5
- Hidden size
hidden_size - 2,048
- Attention heads
num_attention_heads - 64 (8 KV)
- Head dim
head_dim - 128
- Vocabulary
vocab_size - 100,352
- Native dtype
torch_dtype - bfloat16
- Weights on disk
- 924 MB, safetensors
Call it
Create an endpoint for the model, then send it a request in the OpenAI shape.
# Create an endpoint for this model, then provision it.
# --tier-id is the service tier to run on; pick one at /endpoints/new.
erectl endpoints create \
--name "My endpoint" \
--slug my-endpoint \
--model-id 9DE8734B-ED28-453B-8FC5-BE7E7908036F \
--tier-id <tier-id>
# provision takes the endpoint UUID returned by create
erectl endpoints provision <endpoint-id>
# Call it. Requests are routed by /<project-id>/<endpoint-slug>.
curl https://api.erebine.ai/proj_ABC123/my-endpoint/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $EREBINE_API_KEY" \
-d '{
"model": "Laguna-XS-2.1-DFlash",
"messages": [{"role": "user", "content": "Hello, world!"}],
"stream": true
}'Model card
Images are left out; its HTML tables are set as tables.
poolside/Laguna-XS-2.1-DFlash
Model Specifications
| | |
|---|---|
| Base Model | poolside/Laguna-XS-2.1 |
| Draft Architecture | 5-layer Llama-style draft model (DFlashLagunaForCausalLM) |
| Hidden Size | 2048 |
| Attention Heads | 64 query / 8 key-value, head_dim 128 |
| Sliding Window | 512 |
| Gating | Per-head |
| Proposals | Up to 15 tokens per step |
| Chat Template | poolside/Laguna-XS-2.1 (use /chat/completions endpoint) |
| Format | Safetensors (BF16) |
| License | OpenMDW-1.1 |
Deployment
Note. DFlash speculative-decoding support for Laguna XS 2.1 has not yet landed upstream. Integrations are in progress:
- vLLM — vllm-project/vllm#46853
- TRT-LLM — NVIDIA/TensorRT-LLM#15666
See the "Speculative decoding (DFlash)" notes on the Laguna XS 2.1 model card for current status.
Once vLLM support lands, pair this speculator with the base model by adding a --speculative-config block to the standard Laguna XS 2.1 serve command:
# Support in progress — see vllm-project/vllm#46853
vllm serve poolside/Laguna-XS-2.1 \
--tool-call-parser poolside_v1 \
--reasoning-parser poolside_v1 \
--enable-auto-tool-choice \
--default-chat-template-kwargs '{"enable_thinking": true}' \
--speculative-config '{"model": "poolside/Laguna-XS-2.1-DFlash", "num_speculative_tokens": 15, "method": "dflash"}'SGLang support:
sglang serve \
--model-path poolside/Laguna-XS-2.1 \
--speculative-algorithm DFLASH \
--speculative-draft-model-path poolside/Laguna-XS-2.1-DFlashEvaluation
Speculative-decoding throughput and mean acceptance length for the (BF16) Laguna XS 2.1 target paired with this speculator (num_speculative_tokens = 15), versus the same model without speculative decoding:
| Dataset | Baseline (tok/s/seq) | DFlash (tok/s/seq) | Speedup | Acceptance length |
|---|---|---|---|---|
| GSM8K v2 | 80.22 | 133.78 | ×1.67 | 3.55 |
| HumanEval | 95.57 | 252.59 | ×2.64 | 4.57 |
| EvalPlus | 97.68 | 192.69 | ×1.97 | 4.10 |
| Math | 73.63 | 139.84 | ×1.90 | 4.40 |
License
This model is licensed under the OpenMDW-1.1 License.
Intended and Responsible Use
Laguna-XS-2.1-DFlash is designed for software engineering and agentic coding use cases, and you are responsible for confirming that it is appropriate for your intended application. Laguna-XS-2.1-DFlash is subject to the OpenMDW-1.1 License, and should be used consistently with Poolside's Acceptable Use Policy.
Please report security vulnerabilities or safety concerns to security@poolside.ai.