Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75153.49192055716,
-      "required_pages_per_second_upper_estimate": 0.006514114204890027
+      "required_token_positions_per_second": 75271.21976806404,
+      "required_pages_per_second_upper_estimate": 0.0065243185563334215
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9394.186490069646,
-      "required_pages_per_second_upper_estimate": 0.003747481446493396
+      "required_token_positions_per_second": 9408.902471008005,
+      "required_pages_per_second_upper_estimate": 0.0037533518713132297
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18788.37298013929,
-      "required_pages_per_second_upper_estimate": 0.004346359424884355
+      "required_token_positions_per_second": 18817.80494201601,
+      "required_pages_per_second_upper_estimate": 0.004353167991279696
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28182.559470208937,
-      "required_pages_per_second_upper_estimate": 0.0017161700092735972
+      "required_token_positions_per_second": 28226.707413024014,
+      "required_pages_per_second_upper_estimate": 0.0017188583873646805
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28182.559470208937,
-      "required_pages_per_second_upper_estimate": 0.002169571442334436
+      "required_token_positions_per_second": 28226.707413024014,
+      "required_pages_per_second_upper_estimate": 0.002172970073181668
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28182.559470208937,
-      "required_pages_per_second_upper_estimate": 0.3499330676609377
+      "required_token_positions_per_second": 28226.707413024014,
+      "required_pages_per_second_upper_estimate": 0.35048123735704106
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10691251056034068,
-    "producer_p95_seconds": 0.15958742699876893,
-    "gpu_update_seconds": 0.9810322596578211,
-    "producer_batches_per_second": 9.353442312399977,
-    "gpu_batches_per_second": 1.0193344715787376,
-    "producer_headroom": 9.176028647312826,
+    "producer_batch_seconds": 0.04915505401558562,
+    "producer_p95_seconds": 0.17480557599992608,
+    "gpu_update_seconds": 0.9794978775045865,
+    "producer_batches_per_second": 20.34378804024769,
+    "gpu_batches_per_second": 1.0209312577048617,
+    "producer_headroom": 19.926697205825807,
     "sustainable": true,
-    "buffer_seconds": 60.824000098784914,
+    "buffer_seconds": 60.728868405284366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 52,
+    "upload_pages_per_commit": 96,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 760768.6320744833,
-    "upload_bytes_per_second": 49269.77658041964,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 641388.7546368933,
+    "upload_bytes_per_second": 3309682.6814441257,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.62161932790697,
+    "estimated_full_shards_per_hour": 68.81354798951196,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -184,6 +184,12 @@
       "bytes": 43628,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.4819931290003296,
+      "bytes": 7791997,
+      "pages": 45,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -1,2 +1,3 @@
 {"step": 280300, "loss": 0.5914102615788579, "lr": 0.0003, "grad_norm": 0.15056614637032292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41028209266, "tokens_per_second": 75128.25690262954, "data_wait_s": 1.346200042462442e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 280400, "loss": 0.6441070791333914, "lr": 0.0003, "grad_norm": 0.15491111289274895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41042964989, "tokens_per_second": 151113.72276237543, "data_wait_s": 1.8066000848193653e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 280500, "loss": 0.6065780781209469, "lr": 0.0003, "grad_norm": 0.16653364277928817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41057650350, "tokens_per_second": 150594.33820475705, "data_wait_s": 1.4020999515196308e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7b08bcbc276fe18cc4153c7767823eb34f5c0efd808f0829c58c5280d413cfaa
-size 4401
+oid sha256:7dddd06fa7ce288767f5acf838df59ad46362b3407e071d97ca131aa9f09494e
+size 6519