Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75573.46774104232,
-      "required_pages_per_second_upper_estimate": 0.006550516644590695
+      "required_token_positions_per_second": 75789.00710071059,
+      "required_pages_per_second_upper_estimate": 0.006569199050007228
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9446.68346763029,
-      "required_pages_per_second_upper_estimate": 0.0037684232757420975
+      "required_token_positions_per_second": 9473.625887588823,
+      "required_pages_per_second_upper_estimate": 0.0037791710098886325
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18893.36693526058,
-      "required_pages_per_second_upper_estimate": 0.004370647928571166
+      "required_token_positions_per_second": 18947.251775177647,
+      "required_pages_per_second_upper_estimate": 0.004383113237945187
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28340.05040289087,
-      "required_pages_per_second_upper_estimate": 0.0017257603807828623
+      "required_token_positions_per_second": 28420.877662766467,
+      "required_pages_per_second_upper_estimate": 0.0017306823368414272
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28340.05040289087,
-      "required_pages_per_second_upper_estimate": 0.0021816955302950956
+      "required_token_positions_per_second": 28420.877662766467,
+      "required_pages_per_second_upper_estimate": 0.0021879178365044947
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28340.05040289087,
-      "required_pages_per_second_upper_estimate": 0.35188857795660217
+      "required_token_positions_per_second": 28420.877662766467,
+      "required_pages_per_second_upper_estimate": 0.3528921820128198
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.14872579654695528,
+    "producer_batch_seconds": 0.12310137024257983,
     "producer_p95_seconds": 0.38566853500378784,
-    "gpu_update_seconds": 0.9755804808723885,
-    "producer_batches_per_second": 6.723783117774615,
-    "gpu_batches_per_second": 1.0250307582064115,
-    "producer_headroom": 6.559591567320206,
+    "gpu_update_seconds": 0.9728059888953043,
+    "producer_batches_per_second": 8.123386425589173,
+    "gpu_batches_per_second": 1.027954197872051,
+    "producer_headroom": 7.902478964923967,
     "sustainable": true,
-    "buffer_seconds": 60.485989814088086,
+    "buffer_seconds": 60.31397131150887,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328516.1446941511,
-    "upload_bytes_per_second": 5201836.71784731,
+    "generation_bytes_per_second": 313961.3399431058,
+    "upload_bytes_per_second": 5214052.850750632,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.49191718691254,
+    "estimated_full_shards_per_hour": 67.36879490585214,
     "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": 1.1128048520004086,
-      "bytes": 6259334,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2836710900010075,
       "bytes": 2351070,
@@ -562,6 +556,12 @@
       "bytes": 2332449,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9006169920030516,
+      "bytes": 6118351,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -357,3 +357,4 @@
 {"step": 390700, "loss": 0.6517350496724248, "lr": 0.0003, "grad_norm": 0.16227614378862704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57187738408, "tokens_per_second": 149194.12123201502, "data_wait_s": 1.2413998774718493e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 390800, "loss": 0.5534218586981297, "lr": 0.0003, "grad_norm": 0.150613119189268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57202463395, "tokens_per_second": 150439.76599649354, "data_wait_s": 1.170600444311276e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 390900, "loss": 0.5997650064527988, "lr": 0.0003, "grad_norm": 0.14468726891981448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57217087492, "tokens_per_second": 149021.99772401102, "data_wait_s": 7.9809979069978e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 391000, "loss": 0.5779316332191229, "lr": 0.0003, "grad_norm": 0.1475706803031451, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57231675131, "tokens_per_second": 147946.23360693795, "data_wait_s": 1.3527998817153275e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:240d2bacb610f60586122006bb01695890316c2489842bdf5c505049ca6f0fe0
-size 766008
+oid sha256:bfad04259145fdb4930578966f414b3f238bc83a21d1331924ac8f5dc64e60a0
+size 768126