---

license: apache-2.0

base_model: THUDM/glm-edge-1.5b-chat

tags:

- meetara

- gguf

- quantized

- base

- conversational-ai

- emotional-intelligence

- structured-responses

- glm

- thudm

library_name: llama.cpp

quantized_by: meetara

---

meeTARA – GLM‑1.5B‑Base (GGUF, Q4_K_M)

This repository contains a GGUF quantized version of THUDM/glm-edge-1.5b-chat, prepared for use with llama.cpp and compatible runtimes, and used as the core base model inside the meeTARA empathetic assistant.

• Base model: THUDM/glm-edge-1.5b-chat
• Architecture: GLM (1.5B parameters, base‑tuned)
• Format: GGUF
• Quantization: Q4_K_M (good quality vs RAM / speed)
• Intended use: Local GGUF packaging of the upstream model for meeTARA-compatible inference and prompting.

> Note: This repository packages the upstream model as a GGUF for use inside meeTARA. For GLM and Gemma exports, the upstream tokenizer/chat template is preserved. meeTARA's domain routing, emotion analysis, safety layers, and response structuring primarily live in the meeTARA app/runtime prompts rather than inside the GGUF alone.

---

---

Available files

| Filename | Quant type | Size | Notes |

| -------- | ---------- | ---- | ----- |

| meetara-glm-edge-1.5b-chat-gguf-Q4_K_M.gguf | Q4_K_M | ~935MB | Default quant, recommended |

More quantizations (e.g., Q5_K_M, Q8_0) can be added later to this repo as additional .gguf files.

---

Prompt format (recommended)

This export preserves the upstream prompt format for the base model family.

• Use your runtime's native chat templating for THUDM/glm-edge-1.5b-chat when available.
• Do not assume ChatML unless your runtime explicitly requires it.
• For meeTARA-style behavior, pass a system/developer prompt from the meeTARA stack (for example the meetara_default_system_en.txt sidecar produced during conversion).

Practical guidance:

• GLM family: prefer the upstream GLM chat template / automatic conversation mode from your runtime.
• Gemma family: prefer the upstream Gemma turn format / automatic conversation mode from your runtime.

---

Example usage (llama.cpp)

Basic interactive chat

./llama-simple-chat -m /path/to/meetara-glm-edge-1.5b-chat-gguf-Q4_K_M.gguf

With explicit system prompt

./llama-cli \
  -m /path/to/meetara-glm-edge-1.5b-chat-gguf-Q4_K_M.gguf \
  -p "<|im_start|>system You are meeTARA, an emotionally intelligent AI assistant built on top of a GLM‑1.5B‑Base base model. Always answer clearly, kindly, and with practical steps the user can take. <|im_end|> <|im_start|>user How can I improve my sleep quality and manage stress naturally? <|im_end|> <|im_start|>assistant"

Adjust flags like -n (max tokens), --temperature, --top_p, --top_k, etc. according to your hardware and latency/quality trade‑offs.

---

Downloading via huggingface-cli

pip install -U "huggingface_hub[cli]"

huggingface-cli download \
  meetara-glm-edge-1.5b-chat-gguf \
  --include "meetara-glm-edge-1.5b-chat-gguf-Q4_K_M.gguf" \
  --local-dir .

This will download only the Q4_K_M file into the current directory.

---

MeeTARA integration scope

This GGUF is primarily a quantized upstream THUDM/glm-edge-1.5b-chat checkpoint packaged for use inside the meeTARA ecosystem.

In this context, meeTARA packaging means:

* A ready-to-run GGUF build of the upstream model for llama.cpp-style runtimes

* meeTARA naming, repository packaging, and model-card metadata

* Compatibility with meeTARA prompts and orchestration flows

For best results, use this model with the meeTARA runtime or with a meeTARA-style system prompt.

That is where domain routing, emotional adaptation, safety layering, and response structuring are applied most consistently.

When loaded directly in llama.cpp or another GGUF runtime, this repository gives you the upstream model in GGUF form, packaged for clean use inside the meeTARA model family.

---

Intended behavior / meeTARA flavor

Compared to the raw THUDM/glm-edge-1.5b-chat model, this repository is intended to:

* Make the upstream model easy to use inside the meeTARA stack

* Preserve upstream tokenizer/chat-template compatibility

* Provide a clean local-runtime artifact that meeTARA can route to and prompt consistently

The strongest domain detection, emotional intelligence, contextual structure, and safety behavior come from the full meeTARA stack working with this model.

---

Credits

• Base model and original training: THUDM/glm-edge-1.5b-chat by Original Authors.
• Quantization and meeTARA integration: meetara‑lab.

If you use this GGUF in your work, please also cite the original GLM paper/model in addition to this repository.