Update README.md

.eval_results/deepswe.yaml

@@ -1,6 +1,6 @@
 - dataset:
     id: datacurve/deep-swe
-    task_id: deep_swe
+    task_id: deepswe_%_resolved
   value: 40.4
   source:
     url: https://huggingface.co/poolside/Laguna-S-2.1

README.md

@@ -11,37 +11,29 @@ license: openmdw-1.1
 pipeline_tag: text-generation
 ---
 
-<p align="center">
-  <img alt="poolside-banner" src="https://poolside.ai/assets/laguna/laguna-s-2-1-banner.svg" width="800px">
-</p>
-
-<p align="center">
-  <a href="https://openrouter.ai/poolside/laguna-s-2.1"><strong>Use on OpenRouter</strong></a> ·
-  <a href="https://vercel.com/ai-gateway/models/laguna-s-2.1"><strong>Use on Vercel AI Gateway</strong></a> ·
-  <a href="https://poolside.ai/blog/introducing-laguna-s-2-1"><strong>Release blog post</strong></a>
-</p>
-
-<br>
-
 # Laguna S 2.1
 
+> [!IMPORTANT]
+> Private preview. This repository and the rest of the Laguna S 2.1 family are not
+> yet public.
+
 Laguna S 2.1 is a 118B total parameter Mixture-of-Experts model with 8B activated
 parameters per token, designed for agentic coding and long-horizon work. It sits
-between [Laguna XS 2.1](https://huggingface.co/poolside/Laguna-XS-2.1) (33B-A3B) and
-Laguna M.1 (225B-A23B) in the Laguna series and shares the family recipe: a
-token-choice router with softplus gating over 256 routed experts plus one shared
+between [Laguna XS 2.1](https://huggingface.co/poolside/Laguna-XS-2.1) (33B/3B) and
+Laguna M.1 (222B/23B) in the Laguna family and shares the family recipe: a
+token-choice router with sigmoid gating over 256 routed experts plus one shared
 expert, grouped-query attention, and interleaved full/sliding-window attention.
 
 ## Highlights
 
 - **Mixed SWA and global attention layout**: 48 layers in a 1:3 global-to-SWA ratio
-  (12 global attention layers, 36 sliding-window layers, window 512), with softplus
+  (12 global attention layers, 36 sliding-window layers, window 512), with sigmoid
   attention gating and per-layer-type rotary scales
 - **1M context**: 1,048,576-token context window
 - **Native reasoning support**: interleaved thinking between tool calls, with
   per-request control via `enable_thinking`
 - **Speculative decoding**: a trained
-  [DFlash draft model](https://huggingface.co/poolside/Laguna-S-2.1-DFlash) is available
+  [DFlash drafter](https://huggingface.co/poolside/Laguna-S-2.1-DFlash) is available
   for lower-latency serving
 - **Quantized variants**:
   [FP8](https://huggingface.co/poolside/Laguna-S-2.1-FP8),
@@ -57,33 +49,13 @@ expert, grouped-query attention, and interleaved full/sliding-window attention.
 - Number of parameters: 118B total, ~8B activated per token
 - Layers: 48 (12 global attention, 36 sliding-window attention)
 - Experts: 256 routed (top-10) plus 1 shared expert
-- Attention: grouped-query, 8 KV heads, head dim 128; per-head softplus output gating
+- Attention: grouped-query, 8 KV heads, head dim 128; per-head sigmoid output gating
 - Sliding window: 512 tokens
 - Context window: 1,048,576 tokens
 - Vocabulary: 100,352 tokens (Laguna family tokenizer)
 - Modality: text-to-text
 - Reasoning: interleaved thinking with preserved thinking
 
-## Benchmark results
-
-<p align="center">
-  <img alt="benchmarks" src="https://poolside.ai/assets/laguna/laguna-s-2-1-chart.svg" width="800px">
-</p>
-
-| Model | Size | Terminal-Bench 2.1 | SWE-bench Multilingual | SWE-Bench Pro (Public Dataset) | DeepSWE | SWE Atlas (Codebase QnA) | Toolathlon Verified |
-|---|---|---|---|---|---|---|---|
-| **Laguna S 2.1** | 118B-A8B | **70.2%** | **78.5%** | **59.4%** | **40.4%** | **46.2%** | **49.7%** |
-| Tencent Hy3 | 295B-A21B | 71.7% | 75.8% | 57.9% | - | - | - |
-| Inkling | 975B-A41B | 63.8% | - | 54.3% | - | - | 45.5%* |
-| Nemotron 3 Ultra | 550B-A55B | 56.4% | 67.7% | - | - | - | 34.3%* |
-| DeepSeek-V4-Pro Max | 1.6T-A49B | 64.0%* | 76.2% | 55.4% | 9.0%* | 27.2%* | 55.9%* |
-| Kimi K3 | 2800B-A50B | 88.3% | - | - | 69% | - | - |
-| Qwen 3.7 Max | - | 74.5%* | 78.3% | 60.6% | - | - | - |
-| Muse Spark 1.1 | - | 80% | - | 61.5% | 53.3% | 42.2%* | 75.6% |
-| Claude Fable 5 | - | 88% | - | 80.3% | 70% | - | - |
-
-Benchmarks as of 21 July 2026. Laguna S 2.1 in **bold**; a dash (-) marks a benchmark a model was not evaluated on. Scores marked * are as reported by third parties: Terminal-Bench 2.1 and DeepSWE via Artificial Analysis, SWE Atlas via Scale AI's official leaderboard, and Toolathlon Verified via its official leaderboard. Full evaluation trajectories: [trajectories.poolside.ai](https://trajectories.poolside.ai).
-
 ## Usage
 
 Laguna S 2.1 uses the same `laguna` architecture as Laguna XS 2.1, so the same
@@ -106,7 +78,7 @@ vllm serve \
 
 > [!NOTE]
 > **Optional: speculative decoding with DFlash.** Pair with the
-> [Laguna S 2.1 DFlash draft model](https://huggingface.co/poolside/Laguna-S-2.1-DFlash)
+> [Laguna S 2.1 DFlash drafter](https://huggingface.co/poolside/Laguna-S-2.1-DFlash)
 > by adding
 > `--speculative-config '{"model":"poolside/Laguna-S-2.1-DFlash","num_speculative_tokens":7,"method":"dflash"}'`.
 
@@ -150,22 +122,9 @@ cd llama.cpp && cmake -B build && cmake --build build -j
 # with DFlash speculative decoding:
 ./build/bin/llama-server -m laguna-s-2.1-Q4_K_M.gguf \
   -md laguna-s-2.1-DFlash-BF16.gguf \
-  --spec-type draft-dflash --spec-draft-n-max 7 -fa on --jinja --port 8000
+  --spec-type draft-dflash --spec-draft-n-max 15 -fa on --jinja --port 8000
 ```
 
-### Ollama
-
-Run directly from the [Ollama library](https://ollama.com/library/laguna-s-2.1):
-
-```shell
-ollama run laguna-s-2.1
-```
-
-Quantization variants are available as tags (`q4_K_M`, `q8_0`, `f16`, `mxfp8`,
-`nvfp4`, `mlx-bf16`), for example `ollama run laguna-s-2.1:q8_0`. The Laguna chat
-template is baked into the model, so tool-calling and interleaved reasoning work
-automatically.
-
 ## Controlling reasoning
 
 Laguna S 2.1 has native reasoning support and works best with *preserved thinking*:

chat_template.jinja

@@ -1,9 +1,8 @@
 {#- Iteration on laguna_glm_thinking_v8/chat_template.jinja -#}
 {#- No formatting instructions -#}
 {{- "〈|EOS|〉" -}}
-{%- set enable_thinking = enable_thinking | default(true) -%}
+{%- set enable_thinking = enable_thinking | default(false) -%}
 {%- set add_generation_prompt = add_generation_prompt | default(false) -%}
-{%- set preserve_thinking = preserve_thinking | default(false) -%}
 
 {#- ───── header (system message) ───── -#}
 {#- A caller-supplied system message with empty content opts out of the default below, producing no <system> block — used to train without a system message. -#}
@@ -52,7 +51,7 @@
         {%- set reasoning_content = message.reasoning_content -%}
       {%- endif -%}
       {#- Display reasoning content for all messages if enable_thinking -#}
-      {%- if enable_thinking or preserve_thinking -%}
+      {%- if enable_thinking -%}
         {{- '<think>' + reasoning_content + '</think>' -}}
       {%- else -%}
         {{- '</think>' -}}

generation_config.json

@@ -5,6 +5,7 @@
     2,
     24
   ],
+  "max_new_tokens": 32768,
   "pad_token_id": 9,
   "temperature": 1.0,
   "top_p": 1.0,
@@ -21,4 +22,4 @@
     "model": "poolside/Laguna-S-2.1-DFlash",
     "num_speculative_tokens": 15
   }
-}
+}

tokenizer.json

@@ -226,7 +226,7 @@
       "lstrip": false,
       "rstrip": false,
       "normalized": false,
-      "special": true
+      "special": false
     },
     {
       "id": 25,

tokenizer_config.json

@@ -542,7 +542,7 @@
       "lstrip": false,
       "rstrip": false,
       "normalized": false,
-      "special": true
+      "special": false
     },
     "25": {
       "content": "<tool_call>",
@@ -572,5 +572,5 @@
   "sep_token": "〈|SEP|〉",
   "tokenizer_class": "PreTrainedTokenizerFast",
   "unk_token": "〈|UNK|〉",
-  "chat_template": "{#- Iteration on laguna_glm_thinking_v8/chat_template.jinja -#}\n{#- No formatting instructions -#}\n{{- \"〈|EOS|〉\" -}}\n{%- set enable_thinking = enable_thinking | default(true) -%}\n{%- set add_generation_prompt = add_generation_prompt | default(false) -%}\n{%- set preserve_thinking = preserve_thinking | default(false) -%}\n\n{#- ───── header (system message) ───── -#}\n{#- A caller-supplied system message with empty content opts out of the default below, producing no <system> block — used to train without a system message. -#}\n{%- set system_message = \"You are a helpful, conversationally-fluent assistant made by Poolside. You are here to be helpful to users through natural language conversations.\" -%}\n{%- if messages and messages[0].role == \"system\" -%}\n  {%- set system_message = messages[0].content -%}\n  {%- set messages = messages[1:] -%}\n{%- endif -%}\n\n{%- set has_sys = system_message and system_message.strip() -%}\n{%- if has_sys or tools or enable_thinking -%}\n  {{- \"<system>\" -}}\n\n  {%- if has_sys -%}\n    {{- system_message.rstrip() -}}\n    {%- if tools -%}{{- \"\\n\\n\" -}}{%- endif -%}\n  {%- endif -%}\n\n  {%- if tools -%}\n    {{- \"### Tools\\n\\n\" -}}\n    {{- \"You may call functions to assist with the user query.\\n\" -}}\n    {{- \"All available function signatures are listed below:\\n\" -}}\n    {{- \"<available_tools>\\n\" -}}\n    {%- for tool in tools -%}\n      {{- (tool | tojson) ~ \"\\n\" -}}\n    {%- endfor -%}\n    {{- \"</available_tools>\" -}}\n  {%- endif -%}\n\n  {{- \"</system>\\n\" -}}\n{%- endif -%}\n\n{#- ───── main loop ───── -#}\n{%- for message in messages -%}\n  {%- set content = message.content if message.content is string else \"\" -%}\n  {%- if message.role == \"user\" -%}\n    {{- \"<user>\" + content + \"</user>\\n\" -}}\n  {%- elif message.role == \"assistant\" -%}\n    {%- generation -%}\n      {{- \"<assistant>\" -}}\n      {#- Extract reasoning content from message.reasoning (vLLM field name) or message.reasoning_content -#}\n      {%- set reasoning_content = '' -%}\n      {%- if message.reasoning is string -%}\n        {%- set reasoning_content = message.reasoning -%}\n      {%- elif message.reasoning_content is string -%}\n        {%- set reasoning_content = message.reasoning_content -%}\n      {%- endif -%}\n      {#- Display reasoning content for all messages if enable_thinking -#}\n      {%- if enable_thinking or preserve_thinking -%}\n        {{- '<think>' + reasoning_content + '</think>' -}}\n      {%- else -%}\n        {{- '</think>' -}}\n      {%- endif -%}\n      {#- Display main content (trailing newline only when no tool_calls follow) -#}\n      {%- if content -%}\n        {{- content -}}\n      {%- endif -%}\n      {%- if message.tool_calls -%}\n        {%- for tool_call in message.tool_calls -%}\n          {%- set function_data = tool_call.function -%}\n          {{- '<tool_call>' + function_data.name -}}\n          {%- set _args = function_data.arguments -%}\n          {%- for k, v in _args.items() -%}\n            {{- \"<arg_key>\" ~ k ~ \"</arg_key>\" -}}\n            {{- \"<arg_value>\" -}}{{- v | tojson(ensure_ascii=False) if v is not string else v -}}{{- \"</arg_value>\" -}}\n          {%- endfor -%}\n          {{- \"</tool_call>\" -}}\n        {%- endfor -%}\n      {%- endif -%}\n      {{- \"</assistant>\\n\" -}}\n    {%- endgeneration -%}\n  {%- elif message.role == \"tool\" -%}\n    {{- \"<tool_response>\" + content + \"</tool_response>\\n\" -}}\n  {%- elif message.role == \"system\" -%}\n    {#- Render additional system messages (the first one, if any, is handled separately in the header and was sliced off above) -#}\n    {{- \"<system>\" + content + \"</system>\\n\" -}}\n  {%- endif -%}\n{%- endfor -%}\n{#- ───── generation prompt ───── -#}\n{%- if add_generation_prompt -%}\n  {{- \"<assistant>\" -}}\n  {#- ───── Include reasoning mode directive ───── -#}\n  {%- if enable_thinking -%}\n    {{- '<think>' -}}\n  {%- else -%}\n    {{- '</think>' -}}\n  {%- endif -%}\n{%- endif -%}"
+  "chat_template": "{% include 'chat_template.jinja' %}"
 }