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": 77289.75292324387,
-      "required_pages_per_second_upper_estimate": 0.0053470844441646
+      "required_token_positions_per_second": 77438.9937979421,
+      "required_pages_per_second_upper_estimate": 0.005357409274162762
     },
     {
       "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": 9661.219115405484,
-      "required_pages_per_second_upper_estimate": 0.00404495069442348
+      "required_token_positions_per_second": 9679.874224742762,
+      "required_pages_per_second_upper_estimate": 0.004052761199140017
     },
     {
       "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": 19322.438230810967,
-      "required_pages_per_second_upper_estimate": 0.004024354316514884
+      "required_token_positions_per_second": 19359.748449485523,
+      "required_pages_per_second_upper_estimate": 0.004032125051128161
     },
     {
       "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": 28983.657346216452,
-      "required_pages_per_second_upper_estimate": 0.0017375260024233843
+      "required_token_positions_per_second": 29039.622674228285,
+      "required_pages_per_second_upper_estimate": 0.0017408810383835857
     },
     {
       "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": 28983.657346216452,
-      "required_pages_per_second_upper_estimate": 0.0022533518730327605
+      "required_token_positions_per_second": 29039.622674228285,
+      "required_pages_per_second_upper_estimate": 0.002257702931120218
     },
     {
       "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": 28983.657346216452,
-      "required_pages_per_second_upper_estimate": 0.3717712362106239
+      "required_token_positions_per_second": 29039.622674228285,
+      "required_pages_per_second_upper_estimate": 0.37248909934747226
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09059417811720039,
-    "producer_p95_seconds": 0.19929478099948028,
-    "gpu_update_seconds": 0.9539168804591596,
-    "producer_batches_per_second": 11.03823690200395,
-    "gpu_batches_per_second": 1.0483093658209075,
-    "producer_headroom": 10.529560511328786,
+    "producer_batch_seconds": 0.41260563342186174,
+    "producer_p95_seconds": 0.21735042100044666,
+    "gpu_update_seconds": 0.9520784863550137,
+    "producer_batches_per_second": 2.423621780698197,
+    "gpu_batches_per_second": 1.0503335747333726,
+    "producer_headroom": 2.307478156464182,
     "sustainable": true,
-    "buffer_seconds": 60.09676346892705,
+    "buffer_seconds": 60.933023126720876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319176.56161562,
-    "upload_bytes_per_second": 5518920.010587513,
+    "generation_bytes_per_second": 295576.3563074128,
+    "upload_bytes_per_second": 5564196.732182775,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.48786007262659,
+    "estimated_full_shards_per_hour": 63.423805398147465,
     "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": 1.1226819749999777,
-      "bytes": 7652419,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.3251570170000377,
       "bytes": 1162985,
@@ -562,6 +556,12 @@
       "bytes": 7189815,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.272207456000615,
+      "bytes": 11979936,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 615300, "loss": 0.54539555311203, "lr": 0.0001, "grad_norm": 0.16012991082265346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90061832092, "tokens_per_second": 151159.41338497744, "data_wait_s": 1.398799940943718e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 615400, "loss": 0.6175241470336914, "lr": 0.0001, "grad_norm": 0.179007262244534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90076518245, "tokens_per_second": 152883.73590053653, "data_wait_s": 1.0615000064717606e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 615500, "loss": 0.5639321208000183, "lr": 0.0001, "grad_norm": 0.14570288583714838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90091154328, "tokens_per_second": 151532.79202257452, "data_wait_s": 8.449000233667903e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 615600, "loss": 0.5918445587158203, "lr": 0.0001, "grad_norm": 0.16514112255884794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90105904468, "tokens_per_second": 153801.62871890762, "data_wait_s": 1.455899837310426e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3be5b9c98ebeff8c1f8f23724d55e2b9a815918ee50045f03ecc438036035e63
-size 237612
+oid sha256:f23a3071baec76d4186a2187a9c734c84d3a1645680c0d99f1b8f835ab8f8923
+size 239730