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": 74069.84386057474,
-      "required_pages_per_second_upper_estimate": 0.006420186337531649
+      "required_token_positions_per_second": 74984.28151185418,
+      "required_pages_per_second_upper_estimate": 0.006499447475523512
     },
     {
       "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": 9258.730482571842,
-      "required_pages_per_second_upper_estimate": 0.003693446019854732
+      "required_token_positions_per_second": 9373.035188981772,
+      "required_pages_per_second_upper_estimate": 0.0037390438762493106
     },
     {
       "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": 18517.460965143684,
-      "required_pages_per_second_upper_estimate": 0.004283688698103741
+      "required_token_positions_per_second": 18746.070377963544,
+      "required_pages_per_second_upper_estimate": 0.004336573462371366
     },
     {
       "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": 27776.191447715526,
-      "required_pages_per_second_upper_estimate": 0.0016914243287519891
+      "required_token_positions_per_second": 28119.105566945316,
+      "required_pages_per_second_upper_estimate": 0.001712305999481744
     },
     {
       "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": 27776.191447715526,
-      "required_pages_per_second_upper_estimate": 0.0021382881070642103
+      "required_token_positions_per_second": 28119.105566945316,
+      "required_pages_per_second_upper_estimate": 0.002164686585209556
     },
     {
       "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": 27776.191447715526,
-      "required_pages_per_second_upper_estimate": 0.3448873368478529
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28119.105566945316,
+      "required_pages_per_second_upper_estimate": 0.3491451825489566
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.257550455124985,
-    "producer_p95_seconds": 0.2233278160001646,
-    "gpu_update_seconds": 0.995384844320474,
-    "producer_batches_per_second": 3.88273435399179,
-    "gpu_batches_per_second": 1.0046365540985072,
-    "producer_headroom": 3.864814930485874,
+    "producer_batch_seconds": 0.2362641614374752,
+    "producer_p95_seconds": 0.22667953400014085,
+    "gpu_update_seconds": 0.9832460685556402,
+    "producer_batches_per_second": 4.232550522753065,
+    "gpu_batches_per_second": 1.017039408526668,
+    "producer_headroom": 4.161638661460071,
     "sustainable": true,
-    "buffer_seconds": 60.718475503548916,
+    "buffer_seconds": 60.96125625044969,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 255985.82519487734,
-    "upload_bytes_per_second": 6527318.453581974,
+    "generation_bytes_per_second": 255065.67939509332,
+    "upload_bytes_per_second": 6485569.861812115,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.928599041793255,
+    "estimated_full_shards_per_hour": 54.731157173057554,
     "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": 1.176080047999676,
-      "bytes": 12981710,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.0470021729997825,
       "bytes": 7549840,
@@ -562,6 +556,12 @@
       "bytes": 13436756,
       "pages": 47,
       "shards": 4
+    },
+    {
+      "seconds": 1.2015854260007472,
+      "bytes": 10309364,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 13500, "loss": 0.6960186250507832, "lr": 0.0003, "grad_norm": 0.145830692970031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1975092483, "tokens_per_second": 147651.45616285558, "data_wait_s": 9.116000001085922e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 13600, "loss": 0.6929355002939701, "lr": 0.0003, "grad_norm": 0.14627108368837513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1989608705, "tokens_per_second": 138886.82479895413, "data_wait_s": 1.1251000614720397e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 13700, "loss": 0.6662630550563335, "lr": 0.0003, "grad_norm": 0.16334830063970154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2004188834, "tokens_per_second": 147413.34039166078, "data_wait_s": 8.015000275918283e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 13800, "loss": 0.7241083066910505, "lr": 0.0003, "grad_norm": 0.1441718568705546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2018860478, "tokens_per_second": 149899.95310192794, "data_wait_s": 1.2190999768790789e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:87be4c3fd86c663345b09eb20740d2cfd7e2acd27710f1ee6af11451c762c7fd
-size 287043
+oid sha256:e140079cebec87bcf7a551887ee3e1d5c85d7defb7a733d7146c8f6372e792e6
+size 289126