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": 77463.66204649434,
-      "required_pages_per_second_upper_estimate": 0.005359115880835829
+      "required_token_positions_per_second": 77380.50963551304,
+      "required_pages_per_second_upper_estimate": 0.005353363204103967
     },
     {
       "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": 9682.957755811793,
-      "required_pages_per_second_upper_estimate": 0.00405405220920717
+      "required_token_positions_per_second": 9672.56370443913,
+      "required_pages_per_second_upper_estimate": 0.00404970043694991
     },
     {
       "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": 19365.915511623585,
-      "required_pages_per_second_upper_estimate": 0.004033409487535154
+      "required_token_positions_per_second": 19345.12740887826,
+      "required_pages_per_second_upper_estimate": 0.004029079873952438
     },
     {
       "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": 29048.873267435378,
-      "required_pages_per_second_upper_estimate": 0.001741435597321513
+      "required_token_positions_per_second": 29017.691113317385,
+      "required_pages_per_second_upper_estimate": 0.0017395662747945316
     },
     {
       "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": 29048.873267435378,
-      "required_pages_per_second_upper_estimate": 0.00225842212405301
+      "required_token_positions_per_second": 29017.691113317385,
+      "required_pages_per_second_upper_estimate": 0.002255997848726137
     },
     {
       "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": 29048.873267435378,
-      "required_pages_per_second_upper_estimate": 0.3726077560246197
+      "required_token_positions_per_second": 29017.691113317385,
+      "required_pages_per_second_upper_estimate": 0.37220778483238265
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11109052774230577,
-    "producer_p95_seconds": 0.21304043200143497,
-    "gpu_update_seconds": 0.9517752976324284,
-    "producer_batches_per_second": 9.001667561789587,
-    "gpu_batches_per_second": 1.0506681592677727,
-    "producer_headroom": 8.56756482281046,
+    "producer_batch_seconds": 0.061551384648510066,
+    "producer_p95_seconds": 0.21417669500078773,
+    "gpu_update_seconds": 0.9527980669458301,
+    "producer_batches_per_second": 16.246588207731023,
+    "gpu_batches_per_second": 1.0495403325129262,
+    "producer_headroom": 15.47971783879104,
     "sustainable": true,
-    "buffer_seconds": 60.91361904847542,
+    "buffer_seconds": 60.0262782175873,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348750.083195846,
-    "upload_bytes_per_second": 4719917.870342578,
+    "generation_bytes_per_second": 367034.05464445637,
+    "upload_bytes_per_second": 4590336.284813755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.83364936739478,
+    "estimated_full_shards_per_hour": 78.75696401119488,
     "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": 1.1537458799994056,
-      "bytes": 6763666,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.232765078999364,
       "bytes": 12254761,
@@ -562,6 +556,12 @@
       "bytes": 1078327,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 3.8270818399978452,
+      "bytes": 7309721,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -179,3 +179,4 @@
 {"step": 458100, "loss": 0.6139224767684937, "lr": 0.0003, "grad_norm": 0.16996936809218305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67052086938, "tokens_per_second": 154054.12114908855, "data_wait_s": 1.1897998774657026e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 458200, "loss": 0.5873919725418091, "lr": 0.0003, "grad_norm": 0.15693635068716677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67066827930, "tokens_per_second": 153708.19222548825, "data_wait_s": 1.3896002201363444e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 458300, "loss": 0.5666838884353638, "lr": 0.0003, "grad_norm": 0.1448872141843246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67081351229, "tokens_per_second": 150635.5064481098, "data_wait_s": 1.0539999493630603e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 458400, "loss": 0.551595151424408, "lr": 0.0003, "grad_norm": 0.15819763195714914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67095764374, "tokens_per_second": 149919.4412377618, "data_wait_s": 1.5159002941800281e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94dd0dfd9a444b700c70b0b518defa4b9b0959a6fabbe68196d57088a56d4f08
-size 385872
+oid sha256:33134543094b8aeb514df67d902008c157a4a67e3faa2b31854e85099e582b9e
+size 387990