SmolLM2 Blind Spots

Domain:

natural language processing

Record type:

dataset

Creator:

mir

Host:

HuggingFaceTB/SmolLM2-1.7B Used HuggingFace Transformers with AutoModelForCausalLM on Google Colab (T4 GPU, float16). Greedy decoding (do_sample=False) for reproducibility. View Colab Notebook The model struggled with: multi-step arithmetic, low-resource languages (Yoruba), African geographic knowledge, code generation, and logical reasoning. Fine-tuning Dataset Recommendation

tiny-aya-blind-spots

The evaluation of tiny-aya-base on Hausa mathematical reasoning reveals several systematic blind spo

MildredN/deepbrainz-blind-spots

This dataset captures blind spots of the DeepBrainz-R1-0.6B model: Model Link. I generated 10 divers

Tiny Aya Base Blind Spots

This dataset documents blind spots identified in CohereLabs/tiny-aya-base, a multilingual base langu

Tiny-Aya-Base Blind Spots Evaluation Dataset

CohereLabs/tiny-aya-base Architecture: Cohere2 (Cohere Command R family) Parameters: ~3B Type: Raw

Unveiling Cultural Blind Spots: Analyzing the Limitations of mLLMs in Procedural Text Comprehension

Despite the impressive performance of multilingual large language models (mLLMs) in various natural

Possibilities for distributed leadership in South African schools: Policy ambiguities and blind spots

Against the background of the recently published Policy on the South African Standard for Principals

SmolLM2 Blind Spots

Domain:

Record type:

Visit

Tasks

Languages