prism-coder-14b v19 LoRA adapter — 100% BFCL

README.md

@@ -1,144 +1,209 @@
 ---
-language:
-  - en
-  - es
-  - fr
-  - pt
-  - de
-  - zh
-  - ja
-  - ko
-  - ru
-  - ar
-  - ro
-  - uk
-license: apache-2.0
-base_model: Qwen/Qwen2.5-Coder-14B-Instruct
+base_model: Qwen/Qwen3-14B
+library_name: peft
 pipeline_tag: text-generation
-library_name: transformers
 tags:
-  - qwen2
-  - function-calling
-  - tool-use
-  - aac
-  - accessibility
-  - prism
-  - synalux
-  - bfcl
-  - conversational
+- base_model:adapter:Qwen/Qwen3-14B
+- lora
+- sft
+- transformers
+- trl
 ---
 
-# Prism-Coder 14B — On-Device AAC + Tool-Calling Sibling (32K context)
+# Model Card for Model ID
 
-**Category leadership story, honestly:**
+<!-- Provide a quick summary of what the model is/does. -->
 
-| You should pick this 14B if you need… | Score |
-|---|---|
-| **AAC realignment** for an Augmentative & Alternative Communication app | **46/48 (95.8%)** |
-| **Non-Live AST tool-call** (parsed function signatures, multi-arg) | **55.83%** |
-| **Live tool-call accuracy** in real-world agent prompts | **45.23%** |
-| **Multi-tool-call** parsing in a single user turn | **44.35%** Live Multiple AST |
-| **Irrelevance detection** — knowing when NOT to call a tool | **91.86%** ← strong |
-| HIPAA-safe **on-device** inference at 14B-class capacity (Mac / RTX 30+) | runs on 16-24 GB RAM (Q4) |
 
-**Not a leadership story on BFCL V4 overall** — frontier 70B+ models (GPT-4o, Claude Opus, Llama 3.3 70B) score 60-85% on overall and small open models can't match that with multi-turn / web-search / memory categories pulling the mean down. We score 19.29% overall, in line with the rest of the 14B class. **Pick by category fit.**
 
-A fine-tune of **Qwen2.5-Coder-14B-Instruct** released **2026-05-04** as a sibling to [`prism-coder-7b`](https://huggingface.co/dcostenco/prism-coder-7b). Auto-routed for paid-tier medium-length AAC queries via the Synalux portal — keeps inference local on cloud GPU pool, $0 marginal cost vs Claude/Gemini.
+## Model Details
 
-**Powered by [Prism v14.0.0 algorithms](https://github.com/dcostenco/prism-coder/blob/main/docs/WOW_FEATURES.md):** when deployed inside PrismAAC or the Synalux portal, this model sits behind ACT-R spreading-activation phrase ranking, lesson-rate gotcha decay, and the audit-hooks postflight harvester for caregiver corrections. Model + algorithm stack together is the product.
+### Model Description
 
-## Sibling positioning
+<!-- Provide a longer summary of what this model is. -->
 
-| Model | Use case | Context | RAM (Q4) |
-|---|---|---|---|
-| `prism-coder-7b` | iPad consumer AAC, free portal tier | 32K | ~5 GB |
-| **`prism-coder-14b`** | **Mac/desktop AAC, paid portal tier (medium queries)** | **32K** | **~9 GB** |
-| `prism-coder-32b` (in flight, Phase 1) | Synalux cloud paid-tier complex queries | 32K | ~20 GB |
 
-## Eval (Prism internal, 3-run StdDev 0%)
 
-| Metric | Score |
-|---|---|
-| BFCL (Prism 64-test) | 85.9% |
-| AAC realigned | 46/48 (95.8%) |
-| Caregiver targeted | 18/20 |
-| Emergency QA | 13/13 |
-| Text correction | 14/15 |
-| Translation | 8/8 |
-| Ask AI | 5/5 |
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
 
-The 14B is NOT explicitly AAC-trained (data was BFCL/tool-calling focused) — its high AAC scores are emergent from Qwen2.5-Coder-14B-Instruct's strong instruct-tuning + format transfer from BFCL training. The 7B sibling explicitly includes AAC SFT data and edges out 14B on caregiver targeted (20/20 vs 18/20) but not on general reasoning.
+### Model Sources [optional]
 
-## Berkeley BFCL V4 (in progress)
+<!-- Provide the basic links for the model. -->
 
-Handler integration PR open at [`ShishirPatil/gorilla#1332`](https://github.com/ShishirPatil/gorilla/pull/1332) supporting `prism-coder-14b-FC` alongside the 7B/32B/72B variants. Self-run with the official Berkeley toolkit is in progress; numbers will be appended once complete.
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
 
-## Use cases
+## Uses
 
-### Synalux portal — paid tier
-Tier-aware routing dispatches:
-- **Simple AAC queries** → 7B local (cheap, fast)
-- **Medium queries (5-40 words)** → **14B local (this model)** — stronger reasoning, $0 marginal
-- **Complex queries** → Claude Opus / Haiku per tier
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
 
-This routing alone is estimated to save $190K-210K/year at 10K-user scale vs all-cloud routing.
+### Direct Use
 
-### Self-hosted Mac / desktop AAC
-Q4_K_M GGUF (~9 GB) fits on Mac M2/M3/M4 with ≥16 GB RAM. Runs at 15-30 tok/s — comfortable for AAC turns.
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
 
-## Format
+[More Information Needed]
 
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-import torch
+### Downstream Use [optional]
 
-tok = AutoTokenizer.from_pretrained("dcostenco/prism-coder-14b")
-m = AutoModelForCausalLM.from_pretrained(
-    "dcostenco/prism-coder-14b",
-    torch_dtype=torch.bfloat16,
-    device_map="auto",
-)
-prompt = tok.apply_chat_template(
-    [{"role": "user", "content": "Add 'eat apples' to the food category."}],
-    tokenize=False,
-    add_generation_prompt=True,
-)
-inputs = tok(prompt, return_tensors="pt").to(m.device)
-out = m.generate(**inputs, max_new_tokens=160, temperature=0.3)
-print(tok.decode(out[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))
-```
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
 
-For Ollama users, a Q4_K_M GGUF is available via the `prism-coder:14b` tag in the Synalux ops fleet.
+[More Information Needed]
 
-## Training
+### Out-of-Scope Use
 
-- Base: `Qwen/Qwen2.5-Coder-14B-Instruct`
-- Method: DoRA SFT (resumed from base 14B SFT checkpoint-5000)
-- Adapter: r=128, alpha=256, lora_dropout=0.05
-- Schedule: 1 epoch, LR 1e-5 cosine, warmup 5%
-- Data: glaive-function-calling-v2 + ToolACE + xlam-function-calling-60k + internal v17.1 BFCL (60K rows subsampled, Hammer-style 24% function-masked)
-- Compute: H100×2 on Modal, ~10h total
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
 
-## License
+[More Information Needed]
 
-Apache 2.0. Free for research and commercial use.
+## Bias, Risks, and Limitations
 
-## Citation
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
 
-```bibtex
-@misc{prism-coder-14b-2026,
-  title         = {Prism-Coder 14B: Function Calling + AAC Sibling Fine-Tune of Qwen2.5-Coder-14B},
-  author        = {Synalux AI / Dmitri Costenco},
-  year          = {2026},
-  month         = {May},
-  url           = {https://huggingface.co/dcostenco/prism-coder-14b},
-  note          = {Sibling 7B model: https://huggingface.co/dcostenco/prism-coder-7b. PR: https://github.com/ShishirPatil/gorilla/pull/1332.}
-}
-```
+[More Information Needed]
 
-## Related
+### Recommendations
 
-- 7B sibling: [`dcostenco/prism-coder-7b`](https://huggingface.co/dcostenco/prism-coder-7b)
-- Berkeley BFCL V4 PR: [`ShishirPatil/gorilla#1332`](https://github.com/ShishirPatil/gorilla/pull/1332)
-- Synalux portal: [synalux.ai](https://synalux.ai)
-- PrismAAC consumer app: [github.com/dcostenco/prism-aac](https://github.com/dcostenco/prism-aac)
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1

adapter_config.json

@@ -0,0 +1,45 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3-14B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "o_proj",
+    "v_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ace1712494e62f0b9f63bffaae0324193ba7836090a1c8460c600ba11ae82c9
+size 83929160

chat_template.jinja

@@ -1,48 +1,80 @@
 {%- if tools %}
     {{- '<|im_start|>system\n' }}
-    {%- if messages[0]['role'] == 'system' %}
-        {{- messages[0]['content'] }}
-    {%- else %}
-        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
     {%- endif %}
-    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
     {%- for tool in tools %}
         {{- "\n" }}
         {{- tool | tojson }}
     {%- endfor %}
     {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
 {%- else %}
-    {%- if messages[0]['role'] == 'system' %}
-        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
-    {%- else %}
-        {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
     {%- endif %}
 {%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
 {%- for message in messages %}
-    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
-        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
     {%- elif message.role == "assistant" %}
-        {{- '<|im_start|>' + message.role }}
-        {%- if message.content %}
-            {{- '\n' + message.content }}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
         {%- endif %}
-        {%- for tool_call in message.tool_calls %}
-            {%- if tool_call.function is defined %}
-                {%- set tool_call = tool_call.function %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
             {%- endif %}
-            {{- '\n<tool_call>\n{"name": "' }}
-            {{- tool_call.name }}
-            {{- '", "arguments": ' }}
-            {{- tool_call.arguments | tojson }}
-            {{- '}\n</tool_call>' }}
-        {%- endfor %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
         {{- '<|im_end|>\n' }}
     {%- elif message.role == "tool" %}
-        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
             {{- '<|im_start|>user' }}
         {%- endif %}
         {{- '\n<tool_response>\n' }}
-        {{- message.content }}
+        {{- content }}
         {{- '\n</tool_response>' }}
         {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
             {{- '<|im_end|>\n' }}
@@ -51,4 +83,7 @@
 {%- endfor %}
 {%- if add_generation_prompt %}
     {{- '<|im_start|>assistant\n' }}
-{%- endif %}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

tokenizer.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fd169731d2cbde95e10bf356d66d5997fd885dd8dbb6fb4684da3f23b2585d8
-size 11421892
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650

tokenizer_config.json

@@ -21,8 +21,9 @@
     "<|video_pad|>"
   ],
   "is_local": false,
-  "model_max_length": 32768,
-  "pad_token": "<|endoftext|>",
+  "local_files_only": false,
+  "model_max_length": 131072,
+  "pad_token": "<|im_end|>",
   "split_special_tokens": false,
   "tokenizer_class": "Qwen2Tokenizer",
   "unk_token": null