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": 77054.05683655744,
-      "required_pages_per_second_upper_estimate": 0.005330778441997366
+      "required_token_positions_per_second": 76980.5125291023,
+      "required_pages_per_second_upper_estimate": 0.005325690476161315
     },
     {
       "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": 9631.75710456968,
-      "required_pages_per_second_upper_estimate": 0.004032615565723277
+      "required_token_positions_per_second": 9622.564066137787,
+      "required_pages_per_second_upper_estimate": 0.004028766632504316
     },
     {
       "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": 19263.51420913936,
-      "required_pages_per_second_upper_estimate": 0.004012081996731637
+      "required_token_positions_per_second": 19245.128132275575,
+      "required_pages_per_second_upper_estimate": 0.004008252661794364
     },
     {
       "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": 28895.27131370904,
-      "required_pages_per_second_upper_estimate": 0.0017322273947322222
+      "required_token_positions_per_second": 28867.692198413362,
+      "required_pages_per_second_upper_estimate": 0.001730574068881117
     },
     {
       "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": 28895.27131370904,
-      "required_pages_per_second_upper_estimate": 0.00224648024777438
+      "required_token_positions_per_second": 28867.692198413362,
+      "required_pages_per_second_upper_estimate": 0.002244336092867849
     },
     {
       "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": 28895.27131370904,
-      "required_pages_per_second_upper_estimate": 0.37063751508714665
+      "required_token_positions_per_second": 28867.692198413362,
+      "required_pages_per_second_upper_estimate": 0.3702837598082806
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.4623055284142197,
-    "producer_p95_seconds": 0.35440140499849804,
-    "gpu_update_seconds": 0.9568347602565238,
-    "producer_batches_per_second": 2.1630716886085195,
-    "gpu_batches_per_second": 1.0451125330479254,
-    "producer_headroom": 2.0697021805874067,
+    "producer_batch_seconds": 0.6132141768824226,
+    "producer_p95_seconds": 0.40035685500333784,
+    "gpu_update_seconds": 0.957748884461211,
+    "producer_batches_per_second": 1.6307515998472089,
+    "gpu_batches_per_second": 1.0441150245375204,
+    "producer_headroom": 1.5618505255869994,
     "sustainable": true,
-    "buffer_seconds": 60.280589896161,
+    "buffer_seconds": 60.33817972105629,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244938.95841994314,
-    "upload_bytes_per_second": 3716921.7062337627,
+    "generation_bytes_per_second": 224807.54027227437,
+    "upload_bytes_per_second": 3776244.0747976527,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.55819858978959,
+    "estimated_full_shards_per_hour": 48.23846489072965,
     "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": 1.0500477669993415,
-      "bytes": 2444431,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.1570196089996898,
       "bytes": 6331951,
@@ -562,6 +556,12 @@
       "bytes": 5831979,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 2.100949927000329,
+      "bytes": 13094733,
+      "pages": 28,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 880300, "loss": 0.6008081436157227, "lr": 3e-05, "grad_norm": 0.31065557838755636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128850995000, "tokens_per_second": 151393.7339445974, "data_wait_s": 9.249000868294388e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 880400, "loss": 0.5125455856323242, "lr": 3e-05, "grad_norm": 0.3180909913620703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128865660548, "tokens_per_second": 151568.99002227062, "data_wait_s": 1.75360037246719e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 880500, "loss": 0.561471700668335, "lr": 3e-05, "grad_norm": 0.3201716198123726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128880287924, "tokens_per_second": 118832.45793314082, "data_wait_s": 1.4842997188679874e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 880600, "loss": 0.5296868085861206, "lr": 3e-05, "grad_norm": 0.31415713754224395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128895095996, "tokens_per_second": 153974.0727408594, "data_wait_s": 1.1980999261140823e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b35a37d46a1150ef90a78c76d5c325f48e5e043d74046b190e328964ece85394
-size 811269
+oid sha256:506a838946f18cbd7caa47b4c633e2b2faf2b6f37bdd9012e6a8fc9b6fa2fe23
+size 813387