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": 77553.0150778242,
-      "required_pages_per_second_upper_estimate": 0.005365297530870825
+      "required_token_positions_per_second": 77475.82641756214,
+      "required_pages_per_second_upper_estimate": 0.005359957440251529
     },
     {
       "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": 9694.126884728024,
-      "required_pages_per_second_upper_estimate": 0.004058728490246462
+      "required_token_positions_per_second": 9684.478302195268,
+      "required_pages_per_second_upper_estimate": 0.004054688830225311
     },
     {
       "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": 19388.25376945605,
-      "required_pages_per_second_upper_estimate": 0.004038061957542182
+      "required_token_positions_per_second": 19368.956604390536,
+      "required_pages_per_second_upper_estimate": 0.0040340428669595
     },
     {
       "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": 29082.380654184075,
-      "required_pages_per_second_upper_estimate": 0.00174344431399428
+      "required_token_positions_per_second": 29053.434906585808,
+      "required_pages_per_second_upper_estimate": 0.0017417090606233615
     },
     {
       "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": 29082.380654184075,
-      "required_pages_per_second_upper_estimate": 0.002261027175989303
+      "required_token_positions_per_second": 29053.434906585808,
+      "required_pages_per_second_upper_estimate": 0.002258776771432426
     },
     {
       "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": 29082.380654184075,
-      "required_pages_per_second_upper_estimate": 0.37303755280440315
+      "required_token_positions_per_second": 29053.434906585808,
+      "required_pages_per_second_upper_estimate": 0.3726662678337349
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0907152665235742,
-    "producer_p95_seconds": 0.252667239998118,
-    "gpu_update_seconds": 0.9506787057345765,
-    "producer_batches_per_second": 11.02350286034964,
-    "gpu_batches_per_second": 1.051880087318579,
-    "producer_headroom": 10.479809431938596,
+    "producer_batch_seconds": 0.37908943539065376,
+    "producer_p95_seconds": 0.2681729620016995,
+    "gpu_update_seconds": 0.9516258607250868,
+    "producer_batches_per_second": 2.637899942976502,
+    "gpu_batches_per_second": 1.0508331491097296,
+    "producer_headroom": 2.5102938037416713,
     "sustainable": true,
-    "buffer_seconds": 60.843437167012894,
+    "buffer_seconds": 60.904055086405556,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 475557.1205189748,
-    "upload_bytes_per_second": 3773142.858553486,
+    "generation_bytes_per_second": 409120.35441874125,
+    "upload_bytes_per_second": 3794521.742559691,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.04348766018803,
+    "estimated_full_shards_per_hour": 87.78770422383955,
     "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": 0.993486828996538,
-      "bytes": 5717549,
-      "pages": 19,
-      "shards": 2
-    },
     {
       "seconds": 1.089157542999601,
       "bytes": 7603521,
@@ -562,6 +556,12 @@
       "bytes": 1183171,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.7914613989996724,
+      "bytes": 11162167,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -257,3 +257,4 @@
 {"step": 550700, "loss": 0.6051796674728394, "lr": 0.0003, "grad_norm": 0.19117282867623211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80605073568, "tokens_per_second": 152610.71400702806, "data_wait_s": 1.5298999642254785e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 550800, "loss": 0.5747908353805542, "lr": 0.0003, "grad_norm": 0.15569476481396846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80619799994, "tokens_per_second": 153486.48295172173, "data_wait_s": 1.3980999938212335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 550900, "loss": 0.5462967753410339, "lr": 0.0003, "grad_norm": 0.1775689481987864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80634410765, "tokens_per_second": 152436.04036302684, "data_wait_s": 9.16099816095084e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 551000, "loss": 0.5720579028129578, "lr": 0.0003, "grad_norm": 0.1548219317217928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80648910070, "tokens_per_second": 150628.9901113007, "data_wait_s": 1.7459002265240997e-05, "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 sha256:052513d44002892b78968cc0ad13daa6931b187a62b12449e6c179428952563d
-size 552642
+oid sha256:76dd71071d6f51e69b93253c7b0e365f8f4dcc0a20890dc6f0239b3dab057433
+size 554760