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": 76882.74181453539,
-      "required_pages_per_second_upper_estimate": 0.005318926471268263
+      "required_token_positions_per_second": 76920.38283672978,
+      "required_pages_per_second_upper_estimate": 0.005321530564522876
     },
     {
       "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": 9610.342726816923,
-      "required_pages_per_second_upper_estimate": 0.004023649812941256
+      "required_token_positions_per_second": 9615.047854591223,
+      "required_pages_per_second_upper_estimate": 0.004025619751686114
     },
     {
       "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": 19220.685453633847,
-      "required_pages_per_second_upper_estimate": 0.004003161896430075
+      "required_token_positions_per_second": 19230.095709182446,
+      "required_pages_per_second_upper_estimate": 0.00400512180449572
     },
     {
       "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": 28831.028180450772,
-      "required_pages_per_second_upper_estimate": 0.0017283761169869737
+      "required_token_positions_per_second": 28845.143563773665,
+      "required_pages_per_second_upper_estimate": 0.0017292223126642388
     },
     {
       "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": 28831.028180450772,
-      "required_pages_per_second_upper_estimate": 0.002241485626739229
+      "required_token_positions_per_second": 28845.143563773665,
+      "required_pages_per_second_upper_estimate": 0.002242583035705574
     },
     {
       "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": 28831.028180450772,
-      "required_pages_per_second_upper_estimate": 0.36981347315261187
+      "required_token_positions_per_second": 28845.143563773665,
+      "required_pages_per_second_upper_estimate": 0.3699945301339601
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06674609317187219,
-    "producer_p95_seconds": 0.18033385900071153,
-    "gpu_update_seconds": 0.9589668404107441,
-    "producer_batches_per_second": 14.98215030241523,
-    "gpu_batches_per_second": 1.0427889243507946,
-    "producer_headroom": 14.367385338066006,
+    "producer_batch_seconds": 0.36212458675000647,
+    "producer_p95_seconds": 0.18632618299943715,
+    "gpu_update_seconds": 0.9584975695778076,
+    "producer_batches_per_second": 2.7614805417516495,
+    "gpu_batches_per_second": 1.0432994633888044,
+    "producer_headroom": 2.6468723877053635,
     "sustainable": true,
-    "buffer_seconds": 60.414910945876876,
+    "buffer_seconds": 60.38534688340188,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391406.95234401204,
-    "upload_bytes_per_second": 3937205.9001927166,
+    "generation_bytes_per_second": 339984.4707137963,
+    "upload_bytes_per_second": 3968131.725262237,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.98682048549911,
+    "estimated_full_shards_per_hour": 72.9527529817621,
     "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": 7323521,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.158046825999918,
+      "bytes": 6457008,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 531000, "loss": 0.5705332159996033, "lr": 0.0003, "grad_norm": 0.1550770650128346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77721076278, "tokens_per_second": 154057.69925950665, "data_wait_s": 1.1037999684049282e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531100, "loss": 0.555992066860199, "lr": 0.0003, "grad_norm": 0.1407968699878935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77735763717, "tokens_per_second": 153614.5171548599, "data_wait_s": 8.940000043367036e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 531200, "loss": 0.5846951603889465, "lr": 0.0003, "grad_norm": 0.16805864125753955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77750483592, "tokens_per_second": 153595.62734738825, "data_wait_s": 1.3156999557395466e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 531300, "loss": 0.7076463103294373, "lr": 0.0003, "grad_norm": 0.19203421174432692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77765064236, "tokens_per_second": 152195.67498223743, "data_wait_s": 1.3093999768898357e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6ed8644bcd052b6bedd6451ad7768710cac5ca48e0bcd373bf258fea11d733a
-size 132264
+oid sha256:16352065b82009d9f0a4b0ce0e351fd19db71ca67cf43e840d4408d8c368704a
+size 134382