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": 76000.80773194997,
-      "required_pages_per_second_upper_estimate": 0.005257912225065998
+      "required_token_positions_per_second": 76744.91468845106,
+      "required_pages_per_second_upper_estimate": 0.0053093912708827564
     },
     {
       "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": 9500.100966493746,
-      "required_pages_per_second_upper_estimate": 0.003977493890003679
+      "required_token_positions_per_second": 9593.114336056382,
+      "required_pages_per_second_upper_estimate": 0.004016436645499527
     },
     {
       "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": 19000.201932987493,
-      "required_pages_per_second_upper_estimate": 0.0039572409936208894
+      "required_token_positions_per_second": 19186.228672112764,
+      "required_pages_per_second_upper_estimate": 0.003995985457525668
     },
     {
       "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": 28500.30289948124,
-      "required_pages_per_second_upper_estimate": 0.0017085496413811146
+      "required_token_positions_per_second": 28779.343008169144,
+      "required_pages_per_second_upper_estimate": 0.0017252776698273786
     },
     {
       "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": 28500.30289948124,
-      "required_pages_per_second_upper_estimate": 0.0022157731908506208
+      "required_token_positions_per_second": 28779.343008169144,
+      "required_pages_per_second_upper_estimate": 0.0022374673319333796
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28500.30289948124,
-      "required_pages_per_second_upper_estimate": 0.3655712843534747
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28779.343008169144,
+      "required_pages_per_second_upper_estimate": 0.3691505112577974
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08881565930462898,
-    "producer_p95_seconds": 0.21760940199965262,
-    "gpu_update_seconds": 0.9700949529383158,
-    "producer_batches_per_second": 11.25927576093421,
-    "gpu_batches_per_second": 1.03082692778795,
-    "producer_headroom": 10.922566589422992,
+    "producer_batch_seconds": 0.10606299099194416,
+    "producer_p95_seconds": 0.22430119099954027,
+    "gpu_update_seconds": 0.9606890606276868,
+    "producer_batches_per_second": 9.428359417810057,
+    "gpu_batches_per_second": 1.0409195243116733,
+    "producer_headroom": 9.057721752356148,
     "sustainable": true,
-    "buffer_seconds": 60.145887082175584,
+    "buffer_seconds": 60.52341081954427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299614.65129680425,
-    "upload_bytes_per_second": 5496677.509896699,
+    "generation_bytes_per_second": 300590.91360053286,
+    "upload_bytes_per_second": 5566798.772554718,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.29032949617478,
+    "estimated_full_shards_per_hour": 64.49981266033163,
     "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": 0.7514245589991333,
-      "bytes": 257606,
-      "pages": 8,
-      "shards": 1
-    },
     {
       "seconds": 0.9524619530002383,
       "bytes": 4638620,
@@ -562,6 +556,12 @@
       "bytes": 5717549,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.089157542999601,
+      "bytes": 7603521,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -179,3 +179,4 @@
 {"step": 542900, "loss": 0.5737481713294983, "lr": 0.0003, "grad_norm": 0.17190206285617662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79462791169, "tokens_per_second": 131299.12978092476, "data_wait_s": 9.057999704964459e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 543000, "loss": 0.6021584868431091, "lr": 0.0003, "grad_norm": 0.15515566079812285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79477365296, "tokens_per_second": 151846.33496479577, "data_wait_s": 1.1538999387994409e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 543100, "loss": 0.6443955898284912, "lr": 0.0003, "grad_norm": 0.21050887194058637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79492136196, "tokens_per_second": 154191.7396083871, "data_wait_s": 1.8157999875256792e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 543200, "loss": 0.5875047445297241, "lr": 0.0003, "grad_norm": 0.19520319093406965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79506905431, "tokens_per_second": 154025.21101309868, "data_wait_s": 8.560000424040481e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc5cf2c02bc7fd18ceab258979195982e6dc60640ec6dd5d5ff109a92382a1a8
-size 385872
+oid sha256:137eba8c93aa47c50f16f8ef867762885f83db0cea9b4297bd1458d73956a0f4
+size 387990