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": 76152.41167635318,
-      "required_pages_per_second_upper_estimate": 0.005268400537709427
+      "required_token_positions_per_second": 76084.10759480963,
+      "required_pages_per_second_upper_estimate": 0.00526367510286094
     },
     {
       "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": 9519.051459544147,
-      "required_pages_per_second_upper_estimate": 0.0039854280657125865
+      "required_token_positions_per_second": 9510.513449351203,
+      "required_pages_per_second_upper_estimate": 0.003981853379138728
     },
     {
       "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": 19038.102919088295,
-      "required_pages_per_second_upper_estimate": 0.003965134769509469
+      "required_token_positions_per_second": 19021.026898702406,
+      "required_pages_per_second_upper_estimate": 0.0039615782847880286
     },
     {
       "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": 28557.154378632444,
-      "required_pages_per_second_upper_estimate": 0.0017119578007490468
+      "required_token_positions_per_second": 28531.540348053608,
+      "required_pages_per_second_upper_estimate": 0.0017104222787262058
     },
     {
       "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": 28557.154378632444,
-      "required_pages_per_second_upper_estimate": 0.0022201931432913975
+      "required_token_positions_per_second": 28531.540348053608,
+      "required_pages_per_second_upper_estimate": 0.002218201765078107
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28557.154378632444,
-      "required_pages_per_second_upper_estimate": 0.3663005140856639
+      "required_token_positions_per_second": 28531.540348053608,
+      "required_pages_per_second_upper_estimate": 0.3659719648036019
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.23399625666417023,
-    "producer_p95_seconds": 0.32815897099862923,
-    "gpu_update_seconds": 0.9681636914316397,
-    "producer_batches_per_second": 4.273572638536662,
-    "gpu_batches_per_second": 1.0328831878845646,
-    "producer_headroom": 4.1375178613269075,
+    "producer_batch_seconds": 0.0892404978517618,
+    "producer_p95_seconds": 0.3170936520000396,
+    "gpu_update_seconds": 0.9690328549641771,
+    "producer_batches_per_second": 11.205674823342077,
+    "gpu_batches_per_second": 1.0319567544868926,
+    "producer_headroom": 10.858667065863374,
     "sustainable": true,
-    "buffer_seconds": 60.02614886876166,
+    "buffer_seconds": 60.080037007778984,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317382.6369178121,
-    "upload_bytes_per_second": 5761916.167591602,
+    "generation_bytes_per_second": 317433.21362243436,
+    "upload_bytes_per_second": 5719284.803494768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.10292559290669,
+    "estimated_full_shards_per_hour": 68.1137781763532,
     "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": 1.3137797929994122,
-      "bytes": 10288628,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1022804700005508,
       "bytes": 2387118,
@@ -562,6 +556,12 @@
       "bytes": 12708434,
       "pages": 30,
       "shards": 4
+    },
+    {
+      "seconds": 1.1760405560016807,
+      "bytes": 6277751,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -322,3 +322,4 @@
 {"step": 429300, "loss": 0.5749633312225342, "lr": 0.0003, "grad_norm": 0.1391939607394664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62837539403, "tokens_per_second": 151868.3327737305, "data_wait_s": 1.2073996913386509e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429400, "loss": 0.5377009510993958, "lr": 0.0003, "grad_norm": 0.14921407700600425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62852273036, "tokens_per_second": 152181.0550588608, "data_wait_s": 1.119099761126563e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429500, "loss": 0.597822368144989, "lr": 0.0003, "grad_norm": 0.16451327603277427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62866986073, "tokens_per_second": 152906.7777137519, "data_wait_s": 1.8337999790674075e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 429600, "loss": 0.5384544730186462, "lr": 0.0003, "grad_norm": 0.14939019674607884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62881724585, "tokens_per_second": 152498.16433152763, "data_wait_s": 1.6655001672916114e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c087ae3d6675c15a89d8d0e4b9fadfae89fd86b5e814a9017ff418610811880f
-size 691095
+oid sha256:f0f7fd4fe7267edccc743afb59e24f68c86788db568ec51c3e6267573c9d916d
+size 693213