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": 76635.68399314013,
-      "required_pages_per_second_upper_estimate": 0.005301834437931015
+      "required_token_positions_per_second": 76697.9735686161,
+      "required_pages_per_second_upper_estimate": 0.00530614377529417
     },
     {
       "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": 9579.460499142517,
-      "required_pages_per_second_upper_estimate": 0.004010720069108228
+      "required_token_positions_per_second": 9587.246696077013,
+      "required_pages_per_second_upper_estimate": 0.004013979987170421
     },
     {
       "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": 19158.920998285033,
-      "required_pages_per_second_upper_estimate": 0.003990297989219139
+      "required_token_positions_per_second": 19174.493392154025,
+      "required_pages_per_second_upper_estimate": 0.003993541308190407
     },
     {
       "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": 28738.381497427552,
-      "required_pages_per_second_upper_estimate": 0.0017228220898030256
+      "required_token_positions_per_second": 28761.740088231036,
+      "required_pages_per_second_upper_estimate": 0.0017242224016551914
     },
     {
       "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": 28738.381497427552,
-      "required_pages_per_second_upper_estimate": 0.0022342827546438654
+      "required_token_positions_per_second": 28761.740088231036,
+      "required_pages_per_second_upper_estimate": 0.0022360987823352536
     },
     {
       "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": 28738.381497427552,
-      "required_pages_per_second_upper_estimate": 0.3686251009790479
+      "required_token_positions_per_second": 28761.740088231036,
+      "required_pages_per_second_upper_estimate": 0.3689247199013742
     }
   ],
   "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": 0.10202719556258444,
-    "producer_p95_seconds": 0.2820330829999875,
-    "gpu_update_seconds": 0.9620583539986358,
-    "producer_batches_per_second": 9.801308312807546,
-    "gpu_batches_per_second": 1.0394379881882072,
-    "producer_headroom": 9.429430542452774,
+    "producer_batch_seconds": 0.08046427317992766,
+    "producer_p95_seconds": 0.28539275500224903,
+    "gpu_update_seconds": 0.9612770268831278,
+    "producer_batches_per_second": 12.427875881807587,
+    "gpu_batches_per_second": 1.0402828446264119,
+    "producer_headroom": 11.946631578136527,
     "sustainable": true,
-    "buffer_seconds": 60.60967630191406,
+    "buffer_seconds": 60.560452693637046,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 537159.8590870878,
-    "upload_bytes_per_second": 5222417.725986989,
+    "generation_bytes_per_second": 548677.9575523294,
+    "upload_bytes_per_second": 5189939.862355242,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.2620013185451,
+    "estimated_full_shards_per_hour": 117.73351712157643,
     "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": 1.6311224940000102,
-      "bytes": 11525749,
-      "pages": 46,
-      "shards": 4
-    },
     {
       "seconds": 1.0188665900022897,
       "bytes": 6807685,
@@ -562,6 +556,12 @@
       "bytes": 2558602,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.1153945629994269,
+      "bytes": 6138686,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 753000, "loss": 0.5361687541007996, "lr": 5e-05, "grad_norm": 0.24043082826037804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110217965101, "tokens_per_second": 151532.8745609649, "data_wait_s": 1.5124998753890395e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 753100, "loss": 0.5118407607078552, "lr": 5e-05, "grad_norm": 0.23358124361249888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110232619667, "tokens_per_second": 151384.40290076382, "data_wait_s": 1.4532000932376832e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 753200, "loss": 0.5417758226394653, "lr": 5e-05, "grad_norm": 0.23448878078837931, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110247231214, "tokens_per_second": 151290.00648885686, "data_wait_s": 1.2347001757007092e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 753300, "loss": 0.5110217332839966, "lr": 5e-05, "grad_norm": 0.2405861320223855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110261968087, "tokens_per_second": 151740.12718414172, "data_wait_s": 1.4165998436510563e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af91b11334dcaad05cd3a9d4c99ef620903427d8bcbd62e0d9e58c5d24dd630d
-size 708039
+oid sha256:e8aad303a9f99365a6e5a4ae5c0a4805469917931714e25fdc1b1ae47f4c9006
+size 710157