Research: Comprehensive Ecosystem Analysis & Configuration Suite

pmoves/configs/model-suits/glm-4-air.yaml

@@ -1,85 +1,87 @@
-# Model Suit: GLM-4-Air (Z.AI)
-# PMOVES.AI Meta-Agent Configuration
-# Provider: Z.AI (Zhipu AI / BigModel)
+# ============================================================
+# GLM-4-Air Model Suit
+# Role: Lightweight, cost-efficient inference
+# Best for: High-volume tasks, edge deployment, documentation
+# Provider: Zhipu AI (Z.AI)
+# Version: 1.0.0
+# ============================================================
 
-suit:
-  id: glm-4-air
-  name: GLM-4-Air (Zhipu AI)
-  provider: zai
-  flare_name: pmoves/glm-4-air
-  model_family: glm
-  tier: efficient
-  role: balanced
+model_suit:
+  name: glm-4-air
+  provider: zhipu_ai
+  base_url: "https://api.z.ai/v1"
+  api_key_env: ZAI_API_KEY
+
+  architecture:
+    type: dense
+    total_params: "9B"
+    active_params: "9B"
+    attention: standard
+
+  context:
+    max_window: 128000
+    effective_window: 32000
+    working_window: 16000
+
+  defaults:
+    temperature: 0.7
+    top_p: 0.95
+    max_tokens: 4096
+    frequency_penalty: 0.0
+    presence_penalty: 0.0
+
+  advanced:
+    tool_call_parser: glm4
+    enable_thinking: false
+    mtp_steps: 1
+
+  harness_mappings:
+    voice_synthesis:
+      temperature: 0.7
+      top_p: 0.95
+      frequency_penalty: 0.3
+      presence_penalty: 0.1
+      max_tokens: 256
+      system_prompt: conversational_narrator
+
+    documentation:
+      temperature: 0.6
+      top_p: 0.95
+      frequency_penalty: 0.15
+      presence_penalty: 0.1
+      max_tokens: 4096
+      system_prompt: conversational_writer
+
+    lightweight_coding:
+      temperature: 0.6
+      top_p: 0.93
+      frequency_penalty: 0.2
+      presence_penalty: 0.1
+      max_tokens: 2048
+      system_prompt: directive_engineer
+
+    monitoring:
+      temperature: 0.2
+      top_p: 0.85
+      frequency_penalty: 0.0
+      presence_penalty: 0.0
+      max_tokens: 1024
+      system_prompt: directive_infrastructure
+
+    quick_chat:
+      temperature: 0.8
+      top_p: 0.98
+      frequency_penalty: 0.2
+      presence_penalty: 0.1
+      max_tokens: 1024
+      system_prompt: conversational_assistant
 
-model_config:
-  context_window: 128000
-  max_output_tokens: 4096
-  supports_vision: false
-  supports_extended_thinking: false
-  supports_function_calling: true
-  temperature_range: [0.0, 1.0]
-  top_p_range: [0.0, 1.0]
-  top_k: null
-
-distillation_pipeline:
-  stages:
-    - name: config_tuning
-      enabled: true
-
-    - name: context_priming
-      enabled: true
-      system_prompt: |
-        You are GLM-4-Air, Zhipu AI's efficient model for balanced performance.
-        You are a bilingual (Chinese-English) assistant optimized for speed and quality.
-        You prefer instruction-format prompts with clear role separation (system, user, assistant, tool).
-        You have a 128K token context window and can use tools.
-
-    - name: model_fine_tune
-      enabled: false
-      reason: "Cloud-only model - no weight updates possible"
-
-    - name: full_distillation
-      enabled: false
-      reason: "Cloud-only model - no weight updates possible"
-
-tensorzero_config:
-  model_name: glm-4-air
-  weight: 0.0  # Need to register in TensorZero
-  routing:
-    - function: general_chat
-      priority: 3
-    - function: bilingual_chat
-      priority: 3
-
-flare_namespace:
-  alias: pmoves/glm-4-air
-  canonical_name: glm-4-air
-  provider_url: https://open.bigmodel.cn/dev/api
-  already_registered: false  # Need to add to flare-model-namespace.yaml
-
-capabilities:
-  - code_generation
-  - reasoning
-  - tools
-  - function_calling
-  - large_context
-  - bilingual
-
-local_fallback:
-  available: false
-  reason: "Z.AI models are cloud-only with no open weights"
-
-cross_agent:
-  agent_zero: full
-  clawz: full
-  archon: full
-  typer: untested
-  pinokio: full
-  a2ui: full
-
-metadata:
-  source: pmoves/meta-agent/zai
-  last_verified: 2026-04-21
-  benchmark_data:
-    - source: "Z.AI Documentation"
-      url: "https://open.bigmodel.cn/dev/api"
+  cgp_state_vector:
+    delta: 0.4
+    Hz: 0.1
+    kappa: 1.5
+    A: 0.4
+    F: 0.05
+
+  fallback_to: glm-4-flash
+  priority: 10

pmoves/configs/model-suits/glm-4-flash.yaml

@@ -1,86 +1,65 @@
-# Model Suit: GLM-4-Flash (Z.AI)
-# PMOVES.AI Meta-Agent Configuration
-# Provider: Z.AI (Zhipu AI / BigModel)
+# ============================================================
+# GLM-4-Flash Model Suit
+# Role: Fastest inference, lowest latency
+# Best for: Real-time responses, streaming, quick tasks
+# Provider: Zhipu AI (Z.AI)
+# Version: 1.0.0
+# ============================================================
 
-suit:
-  id: glm-4-flash
-  name: GLM-4-Flash (Zhipu AI)
-  provider: zai
-  flare_name: pmoves/glm-4-flash
-  model_family: glm
-  tier: lightning
-  role: fast
+model_suit:
+  name: glm-4-flash
+  provider: zhipu_ai
+  base_url: "https://api.z.ai/v1"
+  api_key_env: ZAI_API_KEY
+
+  architecture:
+    type: dense
+    total_params: "9B"
+    active_params: "9B"
+    attention: standard
+
+  context:
+    max_window: 128000
+    effective_window: 32000
+    working_window: 8000
+
+  defaults:
+    temperature: 0.7
+    top_p: 0.95
+    max_tokens: 4096
+    frequency_penalty: 0.0
+    presence_penalty: 0.0
+
+  advanced:
+    tool_call_parser: glm4
+    enable_thinking: false
+    mtp_steps: 1
+
+  harness_mappings:
+    quick_chat:
+      temperature: 0.8
+      top_p: 0.98
+      max_tokens: 1024
+      system_prompt: conversational_assistant
+
+    streaming_response:
+      temperature: 0.7
+      top_p: 0.95
+      max_tokens: 2048
+      system_prompt: conversational_narrator
+
+    lightweight_coding:
+      temperature: 0.6
+      top_p: 0.93
+      max_tokens: 2048
+      system_prompt: directive_engineer
 
-model_config:
-  context_window: 128000
-  max_output_tokens: 4096
-  supports_vision: false
-  supports_extended_thinking: false
-  supports_function_calling: true
-  temperature_range: [0.0, 1.0]
-  top_p_range: [0.0, 1.0]
-  top_k: null
-
-distillation_pipeline:
-  stages:
-    - name: config_tuning
-      enabled: true
-
-    - name: context_priming
-      enabled: true
-      system_prompt: |
-        You are GLM-4-Flash, Zhipu AI's fastest model for quick responses.
-        You are a bilingual (Chinese-English) assistant optimized for speed.
-        You prefer instruction-format prompts with clear role separation (system, user, assistant, tool).
-        You have a 128K token context window and can use tools.
-
-    - name: model_fine_tune
-      enabled: false
-      reason: "Cloud-only model - no weight updates possible"
-
-    - name: full_distillation
-      enabled: false
-      reason: "Cloud-only model - no weight updates possible"
-
-tensorzero_config:
-  model_name: glm-4-flash
-  weight: 0.0  # Need to register in TensorZero
-  routing:
-    - function: quick_response
-      priority: 4
-    - function: simple_tasks
-      priority: 4
-
-flare_namespace:
-  alias: pmoves/glm-4-flash
-  canonical_name: glm-4-flash
-  provider_url: https://open.bigmodel.cn/dev/api
-  already_registered: false  # Need to add to flare-model-namespace.yaml
-
-capabilities:
-  - code_generation
-  - reasoning
-  - tools
-  - function_calling
-  - large_context
-  - bilingual
-  - fast_response
-
-local_fallback:
-  available: false
-  reason: "Z.AI models are cloud-only with no open weights"
-
-cross_agent:
-  agent_zero: full
-  clawz: full
-  archon: full
-  typer: untested
-  pinokio: full
-  a2ui: full
-
-metadata:
-  source: pmoves/meta-agent/zai
-  last_verified: 2026-04-21
-  benchmark_data:
-    - source: "Z.AI Documentation"
-      url: "https://open.bigmodel.cn/dev/api"
+  cgp_state_vector:
+    delta: 0.5
+    Hz: 0.2
+    kappa: 1.2
+    A: 0.5
+    F: 0.1
+
+  fallback_to: glm-4-air
+  priority: 9