niqun123123 commited on
Commit
05ec18f
·
verified ·
1 Parent(s): 58f04b3

Update model card (2026-11-11)

Browse files
Files changed (1) hide show
  1. README.md +5 -5
README.md CHANGED
@@ -13,7 +13,7 @@ tags:
13
 
14
  # LemonDecider 4B
15
 
16
- **Model ID:** `lemondecider-4b-20261023-v3`
17
  **Base model:** `Qwen/Qwen3.5-4B`
18
 
19
  This v3 release makes typed decisions from a shared text state and named questions. It returns choice probabilities, yes/no probabilities (noul), or ordinal scores without generating an answer token stream. Requests that fit use one joint forward; overlength question lists are recursively bisected, preserving the complete state and every option in each chunk. Inputs are never truncated.
@@ -24,7 +24,7 @@ This v3 release makes typed decisions from a shared text state and named questio
24
  |---|---|
25
  | Architecture | Qwen3.5-4B hybrid backbone with a 255-option decision readout |
26
  | Loader | `modeling_lemondecider.LemonDeciderForDecisionClassification`, `trust_remote_code=True` |
27
- | Model type / protocol | `lemondecider_decision` / `autojev-v1` (functional public spelling of LemonDecider-v1) |
28
  | Attention | Noncausal periodic full-attention layers; hybrid recurrent layers; SDPA |
29
  | Readout | Native inline joint-question readout (`joint_readout.predict_joint`) |
30
  | Temperature | 1.0026751894886363, inherited from the parent checkpoint; not refit |
@@ -82,10 +82,10 @@ recommended reproducing CLI.
82
 
83
  ## Evaluation results
84
 
85
- ### 1) Decision Index, edition 0.3 — full public suite
86
 
87
  On **2026-11-04**, this model was evaluated over the complete public
88
- 0.3 suite using **16 PPU-ZW810E workers**. The run was joint inference plus
89
  iterative question bisection. The official scorer reported `complete: true`.
90
  The underlying selected file contains **140,620 rows**: **140,178 scoreable**
91
  requests and **442 officially excluded** requests. The benchmark-native summary
@@ -94,7 +94,7 @@ remain in index coverage.
94
 
95
  | Headline | Value |
96
  |---|---:|
97
- | **Decision Index (public 0.3)** | **61.67** |
98
  | Raw index | 70.84 |
99
  | Breadth skill | 60.28 |
100
  | Median / mean / p95 request latency | 49.9 ms / 121.2 ms / 286.9 ms |
 
13
 
14
  # LemonDecider 4B
15
 
16
+ **Model ID:** `lemondecider-4b-20261104-v3`
17
  **Base model:** `Qwen/Qwen3.5-4B`
18
 
19
  This v3 release makes typed decisions from a shared text state and named questions. It returns choice probabilities, yes/no probabilities (noul), or ordinal scores without generating an answer token stream. Requests that fit use one joint forward; overlength question lists are recursively bisected, preserving the complete state and every option in each chunk. Inputs are never truncated.
 
24
  |---|---|
25
  | Architecture | Qwen3.5-4B hybrid backbone with a 255-option decision readout |
26
  | Loader | `modeling_lemondecider.LemonDeciderForDecisionClassification`, `trust_remote_code=True` |
27
+ | Model type / protocol | `lemondecider_decision` / `lemondecider-v1` |
28
  | Attention | Noncausal periodic full-attention layers; hybrid recurrent layers; SDPA |
29
  | Readout | Native inline joint-question readout (`joint_readout.predict_joint`) |
30
  | Temperature | 1.0026751894886363, inherited from the parent checkpoint; not refit |
 
82
 
83
  ## Evaluation results
84
 
85
+ ### 1) Decision Index, edition 0.4 — full public suite
86
 
87
  On **2026-11-04**, this model was evaluated over the complete public
88
+ 0.4 suite using **16 PPU-ZW810E workers**. The run was joint inference plus
89
  iterative question bisection. The official scorer reported `complete: true`.
90
  The underlying selected file contains **140,620 rows**: **140,178 scoreable**
91
  requests and **442 officially excluded** requests. The benchmark-native summary
 
94
 
95
  | Headline | Value |
96
  |---|---:|
97
+ | **Decision Index (public 0.4)** | **61.67** |
98
  | Raw index | 70.84 |
99
  | Breadth skill | 60.28 |
100
  | Median / mean / p95 request latency | 49.9 ms / 121.2 ms / 286.9 ms |