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": 75006.8202901214,
-      "required_pages_per_second_upper_estimate": 0.006501401079699704
+      "required_token_positions_per_second": 74628.74988014875,
+      "required_pages_per_second_upper_estimate": 0.006468630894773969
     },
     {
       "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": 9375.852536265174,
-      "required_pages_per_second_upper_estimate": 0.0037401677582037554
+      "required_token_positions_per_second": 9328.593735018594,
+      "required_pages_per_second_upper_estimate": 0.0037213155158044497
     },
     {
       "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": 18751.70507253035,
-      "required_pages_per_second_upper_estimate": 0.004337876949792159
+      "required_token_positions_per_second": 18657.187470037188,
+      "required_pages_per_second_upper_estimate": 0.004316011965908357
     },
     {
       "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": 28127.557608795523,
-      "required_pages_per_second_upper_estimate": 0.0017128206844859851
+      "required_token_positions_per_second": 27985.78120505578,
+      "required_pages_per_second_upper_estimate": 0.001704187245341538
     },
     {
       "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": 28127.557608795523,
-      "required_pages_per_second_upper_estimate": 0.002165337246787934
+      "required_token_positions_per_second": 27985.78120505578,
+      "required_pages_per_second_upper_estimate": 0.0021544229067658447
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28127.557608795523,
-      "required_pages_per_second_upper_estimate": 0.34925012862157173
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27985.78120505578,
+      "required_pages_per_second_upper_estimate": 0.34748974018222406
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.19223750621094737,
-    "producer_p95_seconds": 0.16281288599975596,
-    "gpu_update_seconds": 0.9829506132219044,
-    "producer_batches_per_second": 5.201898524956276,
-    "gpu_batches_per_second": 1.0173451102718287,
-    "producer_headroom": 5.113209345023891,
+    "producer_batch_seconds": 0.34919488444531765,
+    "producer_p95_seconds": 0.17560051700002077,
+    "gpu_update_seconds": 0.9879302563476499,
+    "producer_batches_per_second": 2.8637303825010516,
+    "gpu_batches_per_second": 1.012217202150455,
+    "producer_headroom": 2.829165890894817,
     "sustainable": true,
-    "buffer_seconds": 60.94293801975807,
+    "buffer_seconds": 60.26374563720664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459326.1492413911,
-    "upload_bytes_per_second": 18634.022848171087,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 290987.732077614,
+    "upload_bytes_per_second": 3425789.0439186757,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.56069906169223,
+    "estimated_full_shards_per_hour": 62.4391934561378,
     "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": 24816,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.540608198999962,
+      "bytes": 9815303,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 198000, "loss": 0.6337952017784119, "lr": 0.0003, "grad_norm": 0.18071502585798865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28979500934, "tokens_per_second": 149136.57135252244, "data_wait_s": 1.6908999896259047e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198100, "loss": 0.5864313617348671, "lr": 0.0003, "grad_norm": 0.13070283264529284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28994197011, "tokens_per_second": 148100.4656571597, "data_wait_s": 1.8413000361761078e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198200, "loss": 0.6315398290753365, "lr": 0.0003, "grad_norm": 0.15816601913565825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29008862048, "tokens_per_second": 147508.96141309646, "data_wait_s": 2.0446999769774266e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 198300, "loss": 0.5880254246294498, "lr": 0.0003, "grad_norm": 0.13477129061986676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29023536881, "tokens_per_second": 148109.66909596414, "data_wait_s": 1.2978000086150132e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ad952c4b7a288932e61ce93361b848978a2c9834f21585ec920acb010b70b2c
-size 67941
+oid sha256:e3f62ae795b51f4c3695ed5e2a22877c27290d36201b39f1d6b064d122e855fd
+size 70059