Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77144.54314927735,
-      "required_pages_per_second_upper_estimate": 0.005337038495068501
+      "required_token_positions_per_second": 77546.39792240928,
+      "required_pages_per_second_upper_estimate": 0.005364839740705305
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9643.067893659669,
-      "required_pages_per_second_upper_estimate": 0.00403735115691912
+      "required_token_positions_per_second": 9693.29974030116,
+      "required_pages_per_second_upper_estimate": 0.004058382182147673
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19286.135787319337,
-      "required_pages_per_second_upper_estimate": 0.0040167934748953915
+      "required_token_positions_per_second": 19386.59948060232,
+      "required_pages_per_second_upper_estimate": 0.00403771741280048
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28929.203680979008,
-      "required_pages_per_second_upper_estimate": 0.001734261588338332
+      "required_token_positions_per_second": 29079.89922090348,
+      "required_pages_per_second_upper_estimate": 0.0017432955558580377
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28929.203680979008,
-      "required_pages_per_second_upper_estimate": 0.00224911834007691
+      "required_token_positions_per_second": 29079.89922090348,
+      "required_pages_per_second_upper_estimate": 0.002260834255466408
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28929.203680979008,
-      "required_pages_per_second_upper_estimate": 0.37107276306074843
+      "required_token_positions_per_second": 29079.89922090348,
+      "required_pages_per_second_upper_estimate": 0.3730057236426352
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11342319024225844,
-    "producer_p95_seconds": 0.33863598400057526,
-    "gpu_update_seconds": 0.9557124456273437,
-    "producer_batches_per_second": 8.816539173903669,
-    "gpu_batches_per_second": 1.0463398322113355,
-    "producer_headroom": 8.426076215860757,
+    "producer_batch_seconds": 0.13331641442948694,
+    "producer_p95_seconds": 0.34906877399771474,
+    "gpu_update_seconds": 0.9507598286353692,
+    "producer_batches_per_second": 7.500951809118112,
+    "gpu_batches_per_second": 1.0517903364042058,
+    "producer_headroom": 7.131603656639299,
     "sustainable": true,
-    "buffer_seconds": 60.209884074522655,
+    "buffer_seconds": 60.84862903266363,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 556548.065335322,
-    "upload_bytes_per_second": 3804439.899609041,
+    "generation_bytes_per_second": 550408.0888667927,
+    "upload_bytes_per_second": 3698903.8537535975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.42225904507394,
+    "estimated_full_shards_per_hour": 118.10476302626454,
     "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": 1.1363400170012028,
-      "bytes": 10297243,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2236347540019779,
       "bytes": 6431007,
@@ -562,6 +556,12 @@
       "bytes": 10539085,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 2.2970791090046987,
+      "bytes": 2343674,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -412,3 +412,4 @@
 {"step": 526200, "loss": 0.543147623538971, "lr": 0.0003, "grad_norm": 0.16020451243368156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77018180283, "tokens_per_second": 151837.3865375966, "data_wait_s": 1.4539000403601676e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526300, "loss": 0.638948380947113, "lr": 0.0003, "grad_norm": 0.16214304313669964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77032790042, "tokens_per_second": 152668.93492124518, "data_wait_s": 1.5463003364857286e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526400, "loss": 0.539824366569519, "lr": 0.0003, "grad_norm": 0.15327712469377483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77047436084, "tokens_per_second": 153171.25112464483, "data_wait_s": 1.7212005332112312e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 526500, "loss": 0.5812841653823853, "lr": 0.0003, "grad_norm": 0.17130374021271788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77062021391, "tokens_per_second": 151421.90567680768, "data_wait_s": 1.4544995792675763e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80e2e371ad4e46d513b4725545339e8da9b05a69395bf8ac4b6f00be434a3c03
-size 883281
+oid sha256:e076739a5e677b9944ed7416d5e5aa94067c782f9666fe0ac85bde3fb0c7d9d9
+size 885399