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": 76545.33439431181,
-      "required_pages_per_second_upper_estimate": 0.005295583843044119
+      "required_token_positions_per_second": 76840.20916102586,
+      "required_pages_per_second_upper_estimate": 0.005315983963609122
     },
     {
       "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": 9568.166799288976,
-      "required_pages_per_second_upper_estimate": 0.004005991632818823
+      "required_token_positions_per_second": 9605.026145128233,
+      "required_pages_per_second_upper_estimate": 0.004021423871211052
     },
     {
       "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": 19136.333598577952,
-      "required_pages_per_second_upper_estimate": 0.003985593629529941
+      "required_token_positions_per_second": 19210.052290256466,
+      "required_pages_per_second_upper_estimate": 0.004000947288913942
     },
     {
       "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": 28704.500397866927,
-      "required_pages_per_second_upper_estimate": 0.0017207909696178097
+      "required_token_positions_per_second": 28815.0784353847,
+      "required_pages_per_second_upper_estimate": 0.0017274199541241133
     },
     {
       "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": 28704.500397866927,
-      "required_pages_per_second_upper_estimate": 0.002231648648180234
+      "required_token_positions_per_second": 28815.0784353847,
+      "required_pages_per_second_upper_estimate": 0.0022402456042158563
     },
     {
       "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": 28704.500397866927,
-      "required_pages_per_second_upper_estimate": 0.3681905106125746
+      "required_token_positions_per_second": 28815.0784353847,
+      "required_pages_per_second_upper_estimate": 0.36960888694840627
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.15295083794521247,
-    "producer_p95_seconds": 0.27993752299880725,
-    "gpu_update_seconds": 0.9631939109469595,
-    "producer_batches_per_second": 6.538048522219953,
-    "gpu_batches_per_second": 1.0382125433256266,
-    "producer_headroom": 6.297408526078026,
+    "producer_batch_seconds": 0.4047214123592937,
+    "producer_p95_seconds": 0.29547248400194803,
+    "gpu_update_seconds": 0.9594976484967143,
+    "producer_batches_per_second": 2.470835417801528,
+    "gpu_batches_per_second": 1.0422120383168656,
+    "producer_headroom": 2.3707607732029627,
     "sustainable": true,
-    "buffer_seconds": 60.681216389658445,
+    "buffer_seconds": 60.448351855293,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320675.116567759,
-    "upload_bytes_per_second": 6157891.941618039,
+    "generation_bytes_per_second": 295846.89629131224,
+    "upload_bytes_per_second": 6221664.027676042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.80941508078173,
+    "estimated_full_shards_per_hour": 63.48185698084379,
     "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": 0.8839036259996647,
-      "bytes": 2483088,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.165388396999333,
       "bytes": 6302666,
@@ -562,6 +556,12 @@
       "bytes": 13505764,
       "pages": 42,
       "shards": 5
+    },
+    {
+      "seconds": 1.3861956760010798,
+      "bytes": 10089467,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 511800, "loss": 0.5815263986587524, "lr": 0.0003, "grad_norm": 0.14792045336126503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74910379444, "tokens_per_second": 151404.08318931167, "data_wait_s": 1.487600093241781e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 511900, "loss": 0.6044458150863647, "lr": 0.0003, "grad_norm": 0.15826686798138512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74925019221, "tokens_per_second": 151923.77474090797, "data_wait_s": 1.386900112265721e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 512000, "loss": 0.5737632513046265, "lr": 0.0003, "grad_norm": 0.17436264043066957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74939702847, "tokens_per_second": 152854.79894645113, "data_wait_s": 1.6078000044217333e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 512100, "loss": 0.6296825408935547, "lr": 0.0003, "grad_norm": 0.19045792368815628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74954321694, "tokens_per_second": 151781.22230199672, "data_wait_s": 1.9504001102177426e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:517875e1f5b5f397dba711e53bc7d71022878b1bc9f35703e835de139b256b6c
-size 575940
+oid sha256:9d567dead30265ad42a170c5a22bb0157819748b3d8ba5656f795321e156aad4
+size 578058