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": 75561.19670235415,
-      "required_pages_per_second_upper_estimate": 0.006549453022057868
+      "required_token_positions_per_second": 75744.26954420994,
+      "required_pages_per_second_upper_estimate": 0.006565321311996051
     },
     {
       "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": 9445.149587794269,
-      "required_pages_per_second_upper_estimate": 0.0037678113881419615
+      "required_token_positions_per_second": 9468.033693026242,
+      "required_pages_per_second_upper_estimate": 0.003776940199866859
     },
     {
       "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": 18890.299175588538,
-      "required_pages_per_second_upper_estimate": 0.0043699382563617654
+      "required_token_positions_per_second": 18936.067386052484,
+      "required_pages_per_second_upper_estimate": 0.004380525926359606
     },
     {
       "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": 28335.448763382807,
-      "required_pages_per_second_upper_estimate": 0.001725480165079758
+      "required_token_positions_per_second": 28404.101079078726,
+      "required_pages_per_second_upper_estimate": 0.001729660730914776
     },
     {
       "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": 28335.448763382807,
-      "required_pages_per_second_upper_estimate": 0.0021813412833476117
+      "required_token_positions_per_second": 28404.101079078726,
+      "required_pages_per_second_upper_estimate": 0.0021866263286516563
     },
     {
       "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": 28335.448763382807,
-      "required_pages_per_second_upper_estimate": 0.3518314409946088
+      "required_token_positions_per_second": 28404.101079078726,
+      "required_pages_per_second_upper_estimate": 0.35268387299103177
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.42135872071875724,
-    "producer_p95_seconds": 0.2143757829999231,
-    "gpu_update_seconds": 0.9757389138558067,
-    "producer_batches_per_second": 2.373274720158139,
-    "gpu_batches_per_second": 1.0248643215922602,
-    "producer_headroom": 2.315696497728546,
+    "producer_batch_seconds": 0.08257898628905025,
+    "producer_p95_seconds": 0.20247437400030321,
+    "gpu_update_seconds": 0.9733805665254572,
+    "producer_batches_per_second": 12.10961825687363,
+    "gpu_batches_per_second": 1.0273474059273267,
+    "producer_headroom": 11.787267079282673,
     "sustainable": true,
-    "buffer_seconds": 60.495812659060014,
+    "buffer_seconds": 60.34959512457835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306879.17931769404,
-    "upload_bytes_per_second": 5770306.727559882,
+    "generation_bytes_per_second": 308857.20398511336,
+    "upload_bytes_per_second": 5902367.451914964,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.8491280999004,
+    "estimated_full_shards_per_hour": 66.27356614747096,
     "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": [
-    {
-      "seconds": 1.001895477999824,
-      "bytes": 1118907,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.186168517999704,
       "bytes": 7790528,
@@ -562,6 +556,12 @@
       "bytes": 7543060,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1867137099998217,
+      "bytes": 11783484,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -105,3 +105,4 @@
 {"step": 365500, "loss": 0.5785230807960033, "lr": 0.0003, "grad_norm": 0.16260728728899426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53498888747, "tokens_per_second": 149437.7973033231, "data_wait_s": 1.3671999113284983e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365600, "loss": 0.5742380302399397, "lr": 0.0003, "grad_norm": 0.15051426151775119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53513487408, "tokens_per_second": 149022.14693153457, "data_wait_s": 1.3954000678495504e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365700, "loss": 0.5581132955849171, "lr": 0.0003, "grad_norm": 0.13958854611713198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53528065903, "tokens_per_second": 148731.41572040573, "data_wait_s": 1.3942999430582859e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 365800, "loss": 0.6179000269621611, "lr": 0.0003, "grad_norm": 0.1617581522578019, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53542720019, "tokens_per_second": 149295.59500863997, "data_wait_s": 8.774999514571391e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95989256df0bb014a80ca451e4ade79c33ca008514d7448d91f879d57bcc89e5
-size 228357
+oid sha256:df48a9d28f850486490c87b27c9f828ec5c78778e924604636811d9a3e90a2bf
+size 230475