diff --git a/README.it.md b/README.it.md index e59eec84..09292357 100644 --- a/README.it.md +++ b/README.it.md @@ -198,7 +198,7 @@ Un container **GLM-5.2 int4** pre-convertito è su Hugging Face — **usa la versione con le teste MTP int8**. Pesa circa **372 GB**, quindi mettilo su un disco che abbia lo spazio, meglio se veloce: -**https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp** +**https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp** > ⚠️ Il mirror originale contiene teste MTP int4 → accettazione dei draft allo 0% > ([#8](https://github.com/JustVugg/colibri/issues/8)). Verifica la tua versione: diff --git a/README.md b/README.md index 4b6d6922..a17696c3 100644 --- a/README.md +++ b/README.md @@ -236,14 +236,18 @@ engine still lives in `c/` — an editable install from the clone, not a wheel). ### 2. Get the model -A pre-converted **GLM-5.2 int4** container is on Hugging Face — **use the -version with the int8 MTP heads**. It is about **372 GB**, so put it on a disk -with the room, ideally a fast one: - -**https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp** - -> ⚠️ The original mirror ships int4 MTP heads → 0% draft acceptance -> ([#8](https://github.com/JustVugg/colibri/issues/8)). Check yours: +A pre-converted **GLM-5.2 int4** container is on Hugging Face — use the +**group-scaled (gs64)** build with the **int8 MTP head**. It is about **372 GB**, +so put it on a disk with the room, ideally a fast one: + +**https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp** + +> ⚠️ Use the **gs64** container above, not the older per-row int4 mirrors +> (`mateogrgic/…`, `jlnsrk/…`): those measure ~9pp worse on quality and are the +> root cause of the think-mode loops and never-terminating generations in +> [#455](https://github.com/JustVugg/colibri/issues/455) — the gs64 container +> cured every failing case there. The MTP head must also be **int8, not int4** +> (int4 → 0% draft acceptance, [#8](https://github.com/JustVugg/colibri/issues/8)): > `ls -l /out-mtp-*` — int8 (correct) is `3527131672 / 5366238584 / 1065950496`. Or convert from the FP8 source yourself — one resumable command that never needs diff --git a/README.zh-CN.md b/README.zh-CN.md index 90b1cfd4..12e57e5b 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -178,7 +178,7 @@ git clone https://github.com/JustVugg/colibri && cd colibri/c Hugging Face 上已有预转换的 **GLM-5.2 int4** 容器——请务必使用 **含 int8 MTP head 的版本**。它约为 **372 GB**，请放在空间足够的磁盘上，最好是快盘： -**https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp** +**https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp** > ⚠️ 原始镜像使用 int4 MTP head → 草稿接受率为 0% >（[#8](https://github.com/JustVugg/colibri/issues/8)）。请检查你的版本： diff --git a/README.zh-TW.md b/README.zh-TW.md index 55d5a3d0..4fe6e632 100644 --- a/README.zh-TW.md +++ b/README.zh-TW.md @@ -178,7 +178,7 @@ git clone https://github.com/JustVugg/colibri && cd colibri/c Hugging Face 上已有預先轉換的 **GLM-5.2 int4** 容器——請務必使用 **含 int8 MTP head 的版本**。它約為 **372 GB**，請放在空間足夠的硬碟上，最好是快碟： -**https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp** +**https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp** > ⚠️ 原始鏡像使用 int4 MTP head → 草稿接受率為 0% >（[#8](https://github.com/JustVugg/colibri/issues/8)）。請檢查你的版本： diff --git a/docker/README.IT.md b/docker/README.IT.md index 302f0180..6acbbe7b 100644 --- a/docker/README.IT.md +++ b/docker/README.IT.md @@ -78,7 +78,7 @@ Il modello GLM 5.2 è circa **360 GB**. Scegli uno di questi metodi: 2. **Scarica il modello** (apri il terminale nella cartella dove lo vuoi salvare): ```bash - hf_download mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp --local-dir . + hf_download mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp --local-dir . ``` **Esempio**: se vuoi salvarlo in `C:\LLM\models\glm-5.2` (Windows): @@ -89,7 +89,7 @@ Il modello GLM 5.2 è circa **360 GB**. Scegli uno di questi metodi: #### Metodo B: Senza Python (solo se necessario) Se sei su Windows e non riesci con Python: -- Scarica manualmente da [Hugging Face](https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp) +- Scarica manualmente da [Hugging Face](https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp) - Decomprimi in una cartella (es. `C:\LLM\models\glm-5.2`) --- @@ -347,7 +347,7 @@ Poi riprova il comando `hf_download`. **Soluzione**: 1. Attendi e riprova il comando `hf_download` -2. Se continua, scarica manualmente da [qui](https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp) +2. Se continua, scarica manualmente da [qui](https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp) 3. Decomprimi il file ZIP nella cartella desiderata --- diff --git a/docker/README.md b/docker/README.md index aed1a574..1e171dc2 100644 --- a/docker/README.md +++ b/docker/README.md @@ -78,7 +78,7 @@ python -m pip install -U huggingface_hub[cli] On Linux, use `python3` instead of `python`. 2. **Download the model** (open the terminal in the folder where you want to save it): ```bash -hf_download mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp --local-dir . +hf_download mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp --local-dir . ``` **Example**: if you want to save it in `C:\LLM\models\glm-5.2` (Windows): @@ -90,7 +90,7 @@ hf_download mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp --local-dir . If you are on Windows and cannot get it to work with Python: -* Download manually from [Hugging Face](https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp) +* Download manually from [Hugging Face](https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp) * Unzip into a folder (e.g., `C:\LLM\models\glm-5.2`) --- @@ -371,7 +371,7 @@ Then retry the `hf_download` command. **Solution**: 1. Wait and retry the `hf_download` command -2. If it continues, download manually from [here](https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp) +2. If it continues, download manually from [here](https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp) 3. Unzip the ZIP file into the desired folder ---