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": 77264.5668897861,
-      "required_pages_per_second_upper_estimate": 0.005345342015930343
+      "required_token_positions_per_second": 77320.91380294831,
+      "required_pages_per_second_upper_estimate": 0.005349240226125757
     },
     {
       "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": 9658.070861223263,
-      "required_pages_per_second_upper_estimate": 0.004043632586140408
+      "required_token_positions_per_second": 9665.11422536854,
+      "required_pages_per_second_upper_estimate": 0.004046581495625866
     },
     {
       "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": 19316.141722446526,
-      "required_pages_per_second_upper_estimate": 0.0040230429198726625
+      "required_token_positions_per_second": 19330.22845073708,
+      "required_pages_per_second_upper_estimate": 0.004025976813883578
     },
     {
       "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": 28974.21258366979,
-      "required_pages_per_second_upper_estimate": 0.0017369598033300292
+      "required_token_positions_per_second": 28995.342676105618,
+      "required_pages_per_second_upper_estimate": 0.001738226520107774
     },
     {
       "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": 28974.21258366979,
-      "required_pages_per_second_upper_estimate": 0.0022526175842878774
+      "required_token_positions_per_second": 28995.342676105618,
+      "required_pages_per_second_upper_estimate": 0.002254260356033309
     },
     {
       "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": 28974.21258366979,
-      "required_pages_per_second_upper_estimate": 0.37165008893767126
+      "required_token_positions_per_second": 28995.342676105618,
+      "required_pages_per_second_upper_estimate": 0.37192112307571246
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3773825263906474,
-    "producer_p95_seconds": 0.17990706099953968,
-    "gpu_update_seconds": 0.9542278300112543,
-    "producer_batches_per_second": 2.6498312191721625,
-    "gpu_batches_per_second": 1.0479677583792602,
-    "producer_headroom": 2.5285426941667293,
+    "producer_batch_seconds": 0.07658877901575067,
+    "producer_p95_seconds": 0.18201887699979125,
+    "gpu_update_seconds": 0.9535324451531338,
+    "producer_batches_per_second": 13.05674294395459,
+    "gpu_batches_per_second": 1.0487320123012738,
+    "producer_headroom": 12.450028025084947,
     "sustainable": true,
-    "buffer_seconds": 60.11635329070902,
+    "buffer_seconds": 60.07254404464743,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330656.6924809623,
-    "upload_bytes_per_second": 4058327.747331671,
+    "generation_bytes_per_second": 329936.25181628275,
+    "upload_bytes_per_second": 4055463.4710037583,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.95122891256001,
+    "estimated_full_shards_per_hour": 70.7966391169201,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 10212337,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.5749179960002948,
+      "bytes": 6216583,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -84,3 +84,4 @@
 {"step": 933400, "loss": 0.5656594038009644, "lr": 3e-05, "grad_norm": 0.33002356563542695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136623381897, "tokens_per_second": 151265.67870126484, "data_wait_s": 1.1965999874519184e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 933500, "loss": 0.5297261476516724, "lr": 3e-05, "grad_norm": 0.32865955088158055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136637958800, "tokens_per_second": 151805.82861281827, "data_wait_s": 9.57200063567143e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 933600, "loss": 0.5904932618141174, "lr": 3e-05, "grad_norm": 0.3322589523926682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136652547151, "tokens_per_second": 152252.65762514182, "data_wait_s": 1.0293999366695061e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 933700, "loss": 0.5050845146179199, "lr": 3e-05, "grad_norm": 0.34105192537751106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136667142011, "tokens_per_second": 152649.665056586, "data_wait_s": 7.546999768237583e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7ba8102ca6727d8faf1a7f1f366a73e8ea4ab7e5c57663d12efc060bd0bda25
-size 183096
+oid sha256:f07e12ea5a3427ea3231d5fedc60fd4f7e72826cb3a6a013618b513268928fd8
+size 185214