Update the model card

README.md

@@ -17,7 +17,13 @@ tags:
 
 # Canopy-M
 
-Canopy-M was trained on **Urdu** and **Sindhi**. The numbers below are the full **development** evaluation (greedy CTC, language id on, no language model). The held-out test split has not been scored.
+<div align="center">
+  <img src="figures/canopy-m-logo.png" alt="Canopy-M Logo" width="220"/>
+</div>
+
+Canopy-M is a 57.9-million-parameter Conformer-CTC model for **Urdu** and **Sindhi**. It is non-autoregressive: a Macaron encoder predicts a token at every frame, and greedy CTC reads that sequence. This card reports the full development evaluation, with the language id on and no language model. The held-out test split has not been scored.
+
+Code is Apache-2.0 and lives at [Proxima-AI-Co/canopy](https://github.com/Proxima-AI-Co/canopy). The weights are this repository.
 
 | | Urdu `urd` | Sindhi `snd` | Macro |
 |---|---:|---:|---:|
@@ -30,61 +36,59 @@ Intervals are 95% speaker-cluster bootstrap. Lower is better. Run `canopy_exp_00
 
 ![Canopy-M versus Whisper on the Urdu and Sindhi development sets](figures/overview_cer_wer.png)
 
-## Model
-
-| | |
-|---|---|
-| Architecture | Conformer-CTC, Metro-style Macaron blocks |
-| Size | 57.9M inference parameters, d=384, 12 layers, 6 heads |
-| Languages | Urdu (id 1), Sindhi (id 2). Id 0 is unconditioned |
-| Conditioning | language embedding and a language-ID head |
-| Tokenizer | character, `char_urd_snd` |
-| Frontend | 16 kHz, 80-bin log-mel, 4× conv subsampling, 25 frames/s |
-| Objective | `(1 − 0.3) · CTC + 0.3 · auxiliary CTC` at layer 6 |
-| Decoder in this table | greedy CTC, no n-gram |
-| Training | 40,000 steps, bf16, RTX 3090, about 9,114 hours of audio seen, ~50/50 Urdu/Sindhi |
-| Code | Apache-2.0 |
-
-One block is half SwiGLU, rotary attention, a squeeze-excite depthwise convolution (kernel 31), and a second half SwiGLU, then RMSNorm. The auxiliary head is dropped at export. Sindhi keeps its own letters: Urdu yeh, kaf, and heh mappings are not applied to Sindhi.
-
 ## Use
 
-Each Canopy variant is its own Hub repository. This one is `canopy-m` (`ProximaAI/Canopy-M`). The same loader accepts a short name or any `org/repo` id, so the next checkpoint does not need a new API. Log in once with a token that can read the repo.
-
-From this snapshot, without GitHub:
-
-```bash
-pip install torch torchaudio soundfile numpy pyyaml huggingface_hub
-hf auth login
-hf download ProximaAI/Canopy-M --local-dir Canopy-M
-python Canopy-M/canopy_m.py clip.wav --model canopy-m --language urd
-python Canopy-M/canopy_m.py clip.wav --model ProximaAI/Canopy-M --language snd --device cpu
-```
+`canopy-m` and `ProximaAI/Canopy-M` load the same checkpoint. A later variant is the same call with another short name or Hub id. Log in with a token that can read this repository.
 
-From GitHub, one install for every variant:
+Install the library once, then point it at any published variant:
 
 ```bash
 pip install torch torchaudio soundfile
 pip install git+https://github.com/Proxima-AI-Co/canopy.git
 hf auth login
 canopy-transcribe clip.wav --model canopy-m --language urd
-canopy-transcribe clip.wav --model ProximaAI/Canopy-M --language snd
+canopy-transcribe clip.wav --model ProximaAI/Canopy-M --language snd --device cpu
 ```
 
 ```python
-from canopy.inference import Canopy
+from canopy import Canopy
 
-asr = Canopy.from_pretrained("canopy-m")                 # ProximaAI/Canopy-M
-asr = Canopy.from_pretrained("ProximaAI/Canopy-M")       # same checkpoint
+asr = Canopy.from_pretrained("canopy-m")
 print(asr.transcribe("clip.wav", language="urd"))
 print(asr.transcribe("clip.wav", language="snd"))
 ```
 
-`language` accepts `urd` / `ur` / `urdu` and `snd` / `sd` / `sindhi` for this checkpoint. A future repo is loaded with its own id, for example `Canopy.from_pretrained("ProximaAI/Canopy-S")`, once that repository exists. If a repository contains more than one `.pt` file, pass `filename=`. Decoding here is greedy CTC. Audio may be any sample rate; it is resampled to 16 kHz. Files longer than 30 seconds are transcribed in chunks.
+The snapshot in this repository also runs on its own:
+
+```bash
+pip install torch torchaudio soundfile numpy pyyaml huggingface_hub
+hf auth login
+hf download ProximaAI/Canopy-M --local-dir Canopy-M
+python Canopy-M/canopy_m.py clip.wav --model canopy-m --language urd
+```
+
+`language` accepts `urd`, `ur`, `urdu`, `snd`, `sd`, and `sindhi`. Decoding in the table above is greedy CTC. Audio at any sample rate is resampled to 16 kHz mono. Files longer than 30 seconds are transcribed in chunks. If a repository ships more than one `.pt` file, pass `filename=`.
+
+## Model
+
+| | |
+|---|---|
+| Architecture | Conformer-CTC, Metro-style Macaron blocks |
+| Size | 57.9M inference parameters, d=384, 12 layers, 6 heads |
+| Languages | Urdu (id 1), Sindhi (id 2). Id 0 is unconditioned |
+| Conditioning | Language embedding and a language-ID head |
+| Tokenizer | Character, `char_urd_snd` |
+| Frontend | 16 kHz, 80-bin log-mel, 4× conv subsampling, 25 frames/s |
+| Objective | `(1 − 0.3) · CTC + 0.3 · auxiliary CTC` at layer 6 |
+| Decoder in this table | Greedy CTC, no n-gram |
+| Training | 40,000 steps, bf16, RTX 3090, about 9,114 hours of audio seen, ~50/50 Urdu/Sindhi |
+| Licence | Apache-2.0 |
+
+Each block is half SwiGLU, rotary attention, a squeeze-excite depthwise convolution (kernel 31), and a second half SwiGLU, closed by RMSNorm. The auxiliary head is dropped at export. Sindhi keeps its own letters. Urdu yeh, kaf, and heh mappings are not applied to Sindhi.
 
 ## Training data
 
-About **186.7 hours** used in training. YouTube Sindhi is train-only and is excluded from the scores.
+About **186.7 hours** were used in training. YouTube Sindhi is train-only and is excluded from the scores.
 
 | Language | Source | Train hours | Dev hours | Labels |
 |---|---|---:|---:|---|
@@ -95,7 +99,7 @@ About **186.7 hours** used in training. YouTube Sindhi is train-only and is excl
 
 ## Benchmark
 
-Same development clips and the same `eval_norm` for every system. Whisper is forced to the `urdu` or `sindhi` token, fp16, greedy, `max_new_tokens=225`.
+Every system was scored on the same development clips with the same `eval_norm`. Whisper is forced to the `urdu` or `sindhi` token, fp16, greedy, `max_new_tokens=225`.
 
 | System | Params | Urdu CER | Urdu WER | Sindhi CER | Sindhi WER | Macro CER |
 |---|---:|---:|---:|---:|---:|---:|
@@ -104,7 +108,7 @@ Same development clips and the same `eval_norm` for every system. Whisper is for
 | Whisper-base | 74M | 0.855 | 1.247 | 0.624 | 1.083 | 0.739 |
 | whisper-small-urdu | 244M | 0.177 | 0.330 | not run | not run | — |
 
-`whisper-small-urdu` is `khawajaaliarshad/whisper-small-urdu`, an Urdu-only fine-tune, so it was not run on Sindhi. A value above 1.0 means the output is longer than the reference, usually a repetition or the wrong script.
+`whisper-small-urdu` is [`khawajaaliarshad/whisper-small-urdu`](https://huggingface.co/khawajaaliarshad/whisper-small-urdu), an Urdu-only fine-tune, so it was not run on Sindhi. A value above 1.0 means the hypothesis is longer than the reference, usually a repetition or the wrong script.
 
 ![Model size against median utterance CER](figures/size_vs_cer.png)
 
@@ -121,7 +125,7 @@ Same development clips and the same `eval_norm` for every system. Whisper is for
 
 ![Urdu development CER and WER by source](figures/urdu_by_source.png)
 
-Paired speaker-cluster test, Δ = whisper-small-urdu minus Canopy-M. Positive means Canopy-M makes fewer errors. “No clear difference” means the 95% interval includes zero.
+Paired speaker-cluster comparison. Δ is whisper-small-urdu minus Canopy-M. A positive value means Canopy-M makes fewer errors. “No clear difference” means the 95% interval includes zero.
 
 | Subset | ΔCER [95% CI] | Verdict |
 |---|---|---|
@@ -144,11 +148,11 @@ whisper-small-urdu was fine-tuned on Common Voice Urdu 24.0 with its own split,
 
 ![Sindhi development CER and WER by source](figures/sindhi_by_source.png)
 
-Against zero-shot Whisper-base, Canopy-M is lower on every Sindhi subset. All Sindhi ΔCER is +0.517 [+0.470, +0.568]. Whisper-base did not train on this Sindhi data. Canopy-M did.
+Against zero-shot Whisper-base, Canopy-M is lower on every Sindhi subset. All-Sindhi ΔCER is +0.517 [+0.470, +0.568]. Whisper-base did not train on this Sindhi data. Canopy-M did.
 
 ### Robustness
 
-Median utterance CER is 0.059 on Urdu and 0.078 on Sindhi. Canopy-M produced no runaway outputs (hypothesis longer than 3× the reference) and no empty outputs. Whisper-tiny runaway rate is 10.3% on Urdu and 9.1% on Sindhi.
+Median utterance CER is 0.059 on Urdu and 0.078 on Sindhi. Canopy-M produced no runaway outputs (a hypothesis longer than 3× the reference) and no empty outputs. Whisper-tiny’s runaway rate is 10.3% on Urdu and 9.1% on Sindhi.
 
 ![Per-utterance CER](figures/utterance_cer_distribution.png)
 
@@ -156,7 +160,7 @@ Median utterance CER is 0.059 on Urdu and 0.078 on Sindhi. Canopy-M produced no
 
 ![Paired differences](figures/paired_differences.png)
 
-Decode real-time factor on an RTX 3090 is about 0.0006 for Canopy-M, 0.0218 for Whisper-tiny, 0.0100 for Whisper-base, and 0.0069 for whisper-small-urdu. Batch sizes differed, so treat the speed column as indicative.
+Decode real-time factor on an RTX 3090 is about 0.0006 for Canopy-M, 0.0218 for Whisper-tiny, 0.0100 for Whisper-base, and 0.0069 for whisper-small-urdu. Batch sizes differed, so treat those figures as indicative.
 
 ## Intended use
 
@@ -180,8 +184,9 @@ Surveillance, speaker identification, and any decision about a person made witho
 ```bibtex
 @software{canopy_m_2026,
   title = {Canopy-M: Conformer-CTC speech recognition for Urdu and Sindhi},
+  author = {{Proxima AI}},
   year = {2026},
-  url = {https://github.com/mwzkhalil/canopy},
+  url = {https://huggingface.co/ProximaAI/Canopy-M},
   license = {Apache-2.0}
 }
 ```