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": 76798.73465921993,
-      "required_pages_per_second_upper_estimate": 0.005313114661340084
+      "required_token_positions_per_second": 77401.69694465365,
+      "required_pages_per_second_upper_estimate": 0.0053548289913116455
     },
     {
       "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": 9599.841832402491,
-      "required_pages_per_second_upper_estimate": 0.004019253307733544
+      "required_token_positions_per_second": 9675.212118081707,
+      "required_pages_per_second_upper_estimate": 0.0040508092724368915
     },
     {
       "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": 19199.683664804983,
-      "required_pages_per_second_upper_estimate": 0.003998787777671439
+      "required_token_positions_per_second": 19350.424236163413,
+      "required_pages_per_second_upper_estimate": 0.004030183063389206
     },
     {
       "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": 28799.525497207476,
-      "required_pages_per_second_upper_estimate": 0.0017264875792283503
+      "required_token_positions_per_second": 29025.63635424512,
+      "required_pages_per_second_upper_estimate": 0.001740042579856469
     },
     {
       "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": 28799.525497207476,
-      "required_pages_per_second_upper_estimate": 0.0022390364316827224
+      "required_token_positions_per_second": 29025.63635424512,
+      "required_pages_per_second_upper_estimate": 0.0022566155562608473
     },
     {
       "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": 28799.525497207476,
-      "required_pages_per_second_upper_estimate": 0.36940939055691274
+      "required_token_positions_per_second": 29025.63635424512,
+      "required_pages_per_second_upper_estimate": 0.37230969785206863
     }
   ],
   "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": 0.3845221620702546,
+    "producer_batch_seconds": 0.1204160568515249,
     "producer_p95_seconds": 0.3271497460009414,
-    "gpu_update_seconds": 0.9600158170203488,
-    "producer_batches_per_second": 2.6006303371853345,
-    "gpu_batches_per_second": 1.0416495043839509,
-    "producer_headroom": 2.496646257920884,
+    "gpu_update_seconds": 0.9525372557751474,
+    "producer_batches_per_second": 8.304540325822307,
+    "gpu_batches_per_second": 1.0498277037849075,
+    "producer_headroom": 7.910384052432828,
     "sustainable": true,
-    "buffer_seconds": 60.48099647228197,
+    "buffer_seconds": 60.00984711383428,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377522.73870767246,
-    "upload_bytes_per_second": 4720521.231228898,
+    "generation_bytes_per_second": 377499.7015818863,
+    "upload_bytes_per_second": 4846565.170725534,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.00759144709234,
+    "estimated_full_shards_per_hour": 81.00264821617547,
     "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": 1.2813600879999285,
-      "bytes": 1187408,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.2242876330019499,
       "bytes": 7412020,
@@ -562,6 +556,12 @@
       "bytes": 720156,
       "pages": 23,
       "shards": 1
+    },
+    {
+      "seconds": 1.5615880700061098,
+      "bytes": 13769456,
+      "pages": 47,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -357,3 +357,4 @@
 {"step": 960700, "loss": 0.53617262840271, "lr": 3e-05, "grad_norm": 0.35134323500284453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140620181073, "tokens_per_second": 152239.77556202843, "data_wait_s": 1.535100454930216e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960800, "loss": 0.543138325214386, "lr": 3e-05, "grad_norm": 0.35077671302823576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140634867460, "tokens_per_second": 153093.35816417274, "data_wait_s": 7.340000593103468e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 960900, "loss": 0.5122169256210327, "lr": 3e-05, "grad_norm": 0.3369775143763825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140649557561, "tokens_per_second": 153493.75666208475, "data_wait_s": 1.2072996469214559e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 961000, "loss": 0.5310725569725037, "lr": 3e-05, "grad_norm": 0.35473942127912833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140664331074, "tokens_per_second": 153975.99668299165, "data_wait_s": 1.0764997568912804e-05, "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 sha256:96e1eb98c277b429a3361e3f7943267cbeb3cf64474fb0bc5b072f7f55a12fbd
-size 766008
+oid sha256:9cf749d32e8b4bbefc8fc16d8806c721072d6a898f0f60e0611cdd12dfeb9f91
+size 768126