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": 73414.15833015913,
-      "required_pages_per_second_upper_estimate": 0.006363353177575008
+      "required_token_positions_per_second": 75116.97839899872,
+      "required_pages_per_second_upper_estimate": 0.006510949305384016
     },
     {
       "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": 9176.769791269891,
-      "required_pages_per_second_upper_estimate": 0.0036607506746728466
+      "required_token_positions_per_second": 9389.62229987484,
+      "required_pages_per_second_upper_estimate": 0.0037456607227839634
     },
     {
       "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": 18353.539582539783,
-      "required_pages_per_second_upper_estimate": 0.004245768371156131
+      "required_token_positions_per_second": 18779.24459974968,
+      "required_pages_per_second_upper_estimate": 0.0043442477347352265
     },
     {
       "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": 27530.309373809672,
-      "required_pages_per_second_upper_estimate": 0.0016764514004946646
+      "required_token_positions_per_second": 28168.866899624518,
+      "required_pages_per_second_upper_estimate": 0.001715336203564372
     },
     {
       "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": 27530.309373809672,
-      "required_pages_per_second_upper_estimate": 0.002119359424369794
+      "required_token_positions_per_second": 28168.866899624518,
+      "required_pages_per_second_upper_estimate": 0.002168517350347386
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27530.309373809672,
-      "required_pages_per_second_upper_estimate": 0.341834304404307
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28168.866899624518,
+      "required_pages_per_second_upper_estimate": 0.34976305176036504
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.5107839147500783,
-    "producer_p95_seconds": 0.21383380999941437,
-    "gpu_update_seconds": 1.0042749474621702,
-    "producer_batches_per_second": 1.957775041700932,
-    "gpu_batches_per_second": 0.9957432499207781,
-    "producer_headroom": 1.9661444271469517,
+    "producer_batch_seconds": 0.08428292574231477,
+    "producer_p95_seconds": 0.21030219300155295,
+    "gpu_update_seconds": 0.9815091284473547,
+    "producer_batches_per_second": 11.864799319584414,
+    "gpu_batches_per_second": 1.018839225246836,
+    "producer_headroom": 11.645408839368065,
     "sustainable": true,
-    "buffer_seconds": 60.25649684773021,
+    "buffer_seconds": 60.853565963735996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266188.485787771,
-    "upload_bytes_per_second": 3582254.083984283,
+    "generation_bytes_per_second": 268096.43718736334,
+    "upload_bytes_per_second": 3661950.9345863685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.117852499245146,
+    "estimated_full_shards_per_hour": 57.52725445476222,
     "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": 0.7696208980000847,
-      "bytes": 1062863,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.0704197729999123,
       "bytes": 6206676,
@@ -562,6 +556,12 @@
       "bytes": 910923,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.2945012599993788,
+      "bytes": 12063808,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 245700, "loss": 0.5668801683932543, "lr": 0.0003, "grad_norm": 0.15620562866162938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35963430168, "tokens_per_second": 148718.40800135015, "data_wait_s": 1.2986000001546927e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 245800, "loss": 0.6115876995027065, "lr": 0.0003, "grad_norm": 0.16180542161259748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35977951781, "tokens_per_second": 148257.8724533782, "data_wait_s": 1.6004998542484827e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 245900, "loss": 0.61958840303123, "lr": 0.0003, "grad_norm": 0.16966736014105382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35992654748, "tokens_per_second": 121703.00145173802, "data_wait_s": 1.2460999641916715e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 246000, "loss": 0.6332357935607433, "lr": 0.0003, "grad_norm": 0.15884492809949485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36007415822, "tokens_per_second": 151095.6949204287, "data_wait_s": 1.1965999874519184e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93d1e7c9a4540910ba2206fa474dfdda9097566e95ca4075ac01229557e4051f
-size 188115
+oid sha256:fc1f1223f89f8b4967206dcc90a303587a161527e3c19049c348ebdb8e7aec9f
+size 190233