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": 75476.97460351055,
-      "required_pages_per_second_upper_estimate": 0.0065421528639890624
+      "required_token_positions_per_second": 75557.32773311304,
+      "required_pages_per_second_upper_estimate": 0.006549117669609861
     },
     {
       "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": 9434.621825438819,
-      "required_pages_per_second_upper_estimate": 0.003763611706334298
+      "required_token_positions_per_second": 9444.66596663913,
+      "required_pages_per_second_upper_estimate": 0.0037676184644323957
     },
     {
       "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": 18869.243650877637,
-      "required_pages_per_second_upper_estimate": 0.004365067431284477
+      "required_token_positions_per_second": 18889.33193327826,
+      "required_pages_per_second_upper_estimate": 0.004369714501876166
     },
     {
       "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": 28303.865476316456,
-      "required_pages_per_second_upper_estimate": 0.001723556908602118
+      "required_token_positions_per_second": 28333.99789991739,
+      "required_pages_per_second_upper_estimate": 0.0017253918151068064
     },
     {
       "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": 28303.865476316456,
-      "required_pages_per_second_upper_estimate": 0.0021789099144811095
+      "required_token_positions_per_second": 28333.99789991739,
+      "required_pages_per_second_upper_estimate": 0.002181229591861797
     },
     {
       "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": 28303.865476316456,
-      "required_pages_per_second_upper_estimate": 0.35143928227170684
+      "required_token_positions_per_second": 28333.99789991739,
+      "required_pages_per_second_upper_estimate": 0.35181342612609595
     }
   ],
   "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": 0.13068091485919808,
-    "producer_p95_seconds": 0.33731671900022775,
-    "gpu_update_seconds": 0.9768277065595419,
-    "producer_batches_per_second": 7.652226808156709,
-    "gpu_batches_per_second": 1.0237219862672329,
-    "producer_headroom": 7.474907163085161,
+    "producer_batch_seconds": 0.1023348968669211,
+    "producer_p95_seconds": 0.3140867639995122,
+    "gpu_update_seconds": 0.9757888772936137,
+    "producer_batches_per_second": 9.771837668439002,
+    "gpu_batches_per_second": 1.0248118453384474,
+    "producer_headroom": 9.535250507581537,
     "sustainable": true,
-    "buffer_seconds": 60.56331780669159,
+    "buffer_seconds": 60.49891039220405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389326.70897125686,
-    "upload_bytes_per_second": 5018595.878727291,
+    "generation_bytes_per_second": 386999.8357331103,
+    "upload_bytes_per_second": 5028062.556453729,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.54044868329315,
+    "estimated_full_shards_per_hour": 83.04115585322363,
     "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": 0.7929018860013457,
-      "bytes": 738445,
-      "pages": 21,
-      "shards": 1
-    },
     {
       "seconds": 1.138339361998078,
       "bytes": 10270327,
@@ -562,6 +556,12 @@
       "bytes": 6328963,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.9812373249988013,
+      "bytes": 2474379,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -311,3 +311,4 @@
 {"step": 386100, "loss": 0.5972938779741526, "lr": 0.0003, "grad_norm": 0.16952418522300677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56514234255, "tokens_per_second": 151120.4774845238, "data_wait_s": 7.779999577905983e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 386200, "loss": 0.6052970960736275, "lr": 0.0003, "grad_norm": 0.1709050477368543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56528815054, "tokens_per_second": 148483.81847558808, "data_wait_s": 1.5281002561096102e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 386300, "loss": 0.5637060552835464, "lr": 0.0003, "grad_norm": 0.16392651726264335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56543365074, "tokens_per_second": 148497.18542614067, "data_wait_s": 1.1593001545406878e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 386400, "loss": 0.5703699178993702, "lr": 0.0003, "grad_norm": 0.18797698578785957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56558064033, "tokens_per_second": 149667.19114507549, "data_wait_s": 1.2093001714674756e-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:c42e28d81a375cc0fc1316871f4cbbdf4bafd72b26ad49f6c403087d4a89ceef
-size 667797
+oid sha256:2118ddae47e27576f30bad82b052e64d1d18a960c99ceab61c75c058849ae147
+size 669915