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": 77169.33957218908,
-      "required_pages_per_second_upper_estimate": 0.005338753969141163
+      "required_token_positions_per_second": 76292.77819935197,
+      "required_pages_per_second_upper_estimate": 0.005278111419465689
     },
     {
       "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": 9646.167446523636,
-      "required_pages_per_second_upper_estimate": 0.004038648874977239
+      "required_token_positions_per_second": 9536.597274918997,
+      "required_pages_per_second_upper_estimate": 0.003992774132211747
     },
     {
       "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": 19292.33489304727,
-      "required_pages_per_second_upper_estimate": 0.004018084585137109
+      "required_token_positions_per_second": 19073.194549837994,
+      "required_pages_per_second_upper_estimate": 0.0039724434307660955
     },
     {
       "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": 28938.502339570907,
-      "required_pages_per_second_upper_estimate": 0.0017348190287226866
+      "required_token_positions_per_second": 28609.79182475699,
+      "required_pages_per_second_upper_estimate": 0.0017151133352714869
     },
     {
       "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": 28938.502339570907,
-      "required_pages_per_second_upper_estimate": 0.002249841269881953
+      "required_token_positions_per_second": 28609.79182475699,
+      "required_pages_per_second_upper_estimate": 0.0022242854731999244
     },
     {
       "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": 28938.502339570907,
-      "required_pages_per_second_upper_estimate": 0.37119203626904357
+      "required_token_positions_per_second": 28609.79182475699,
+      "required_pages_per_second_upper_estimate": 0.3669756907268633
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.10931912659344789,
-    "producer_p95_seconds": 0.32086612199782394,
-    "gpu_update_seconds": 0.9554053515130859,
-    "producer_batches_per_second": 9.14753009067615,
-    "gpu_batches_per_second": 1.0466761552217485,
-    "producer_headroom": 8.73959920175898,
+    "producer_batch_seconds": 0.09154915285955667,
+    "producer_p95_seconds": 0.3261282250023214,
+    "gpu_update_seconds": 0.9663824249177262,
+    "producer_batches_per_second": 10.923093974818922,
+    "gpu_batches_per_second": 1.034787030698676,
+    "producer_headroom": 10.555886042989714,
     "sustainable": true,
-    "buffer_seconds": 60.19053714532441,
+    "buffer_seconds": 60.88209276981675,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384813.33151547407,
-    "upload_bytes_per_second": 5740120.412128498,
+    "generation_bytes_per_second": 336081.2992223133,
+    "upload_bytes_per_second": 5716296.3033110965,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.57198294733206,
+    "estimated_full_shards_per_hour": 72.1152232408719,
     "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": 1.4659404199992423,
-      "bytes": 13390262,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.8803229269979056,
       "bytes": 4831371,
@@ -562,6 +556,12 @@
       "bytes": 2311552,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.171514493998984,
+      "bytes": 9915176,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -326,3 +326,4 @@
 {"step": 429700, "loss": 0.5899538397789001, "lr": 0.0003, "grad_norm": 0.15180337830052834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62896493785, "tokens_per_second": 152997.20772784244, "data_wait_s": 1.1485000868560746e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429800, "loss": 0.5884919166564941, "lr": 0.0003, "grad_norm": 0.18113422230842818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62911194137, "tokens_per_second": 153075.25545081132, "data_wait_s": 9.485996997682378e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 429900, "loss": 0.5997447967529297, "lr": 0.0003, "grad_norm": 0.16632076803459098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62925724703, "tokens_per_second": 143131.93248850532, "data_wait_s": 1.1842999811051413e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 430000, "loss": 0.5564310550689697, "lr": 0.0003, "grad_norm": 0.17130016792381678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62940228981, "tokens_per_second": 149703.11121873936, "data_wait_s": 1.0424002539366484e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f9ea17ccad09585d0c85332d591f47e0c402216c2c4b5de28c296da4b129710
-size 699567
+oid sha256:39cbd482c3235d81ba6839b5d6fb02d518300ba740cec0e88bcd93d2f5764150
+size 701685