初始化项目，由ModelHub XC社区提供模型

Model: beaunix/aegis-geo-mind-qwen2.5-7b
Source: Original Platform

README.md

@@ -0,0 +1,55 @@
+---
+license: apache-2.0
+base_model: Qwen/Qwen2.5-7B-Instruct
+tags:
+  - geology
+  - earth-science
+  - petroleum-geology
+  - qlora
+  - qwen2.5
+language:
+  - en
+---
+ 
+# Aegis-Geo-Mind (Qwen2.5-7B, QLoRA)
+ 
+A Qwen2.5-7B-Instruct model fine-tuned with QLoRA on a curated geology and
+Earth-science corpus, with an emphasis on petroleum geology (sedimentology of
+petroliferous basins, reservoir stratigraphy, well logging, and seismic
+interpretation).
+ 
+## Intended use
+ 
+Educational and research assistant for geology and petroleum-geology concepts.
+It produces fluent, domain-appropriate explanations using correct terminology
+across structural geology, stratigraphy, sedimentology, petroleum systems, and
+well-log interpretation.
+ 
+## Training
+ 
+- Base model: Qwen/Qwen2.5-7B-Instruct
+- Method: QLoRA (4-bit), LoRA rank 16
+- Sequence length: 1024 tokens
+- Corpus: filtered geoscience QA (public GeoSignal subset) combined with
+  curated general-geology and petroleum-geology QA pairs, deduplicated and
+  quality-filtered.
+ 
+## Limitations (please read)
+ 
+This is a supervised fine-tune **without retrieval augmentation**. It reliably
+reproduces the *style and structure* of expert geological reasoning, but it can
+state **specific facts, numbers, and ratios incorrectly with high confidence**.
+Known weak areas include:
+ 
+- Kerogen type classification (H/C and O/C ratios can be inverted)
+- Numeric ranges (oil-window temperatures, burial depths, maturity cutoffs)
+- Occasional invented specific values
+ 
+For any factual or numeric use, verify against authoritative sources
+(e.g. Tissot & Welte, Selley, Schlumberger logging references). A
+retrieval-augmented (RAG) version addressing these gaps is planned.
+ 
+## Not for operational decisions
+ 
+This model must not be used as a sole basis for exploration, drilling, or any
+operational geological decision.