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": 74909.22586642942,
-      "required_pages_per_second_upper_estimate": 0.006492941842404889
+      "required_token_positions_per_second": 75458.94798449054,
+      "required_pages_per_second_upper_estimate": 0.006540590362340457
     },
     {
       "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": 9363.653233303678,
-      "required_pages_per_second_upper_estimate": 0.003735301273856581
+      "required_token_positions_per_second": 9432.368498061318,
+      "required_pages_per_second_upper_estimate": 0.0037627128203531666
     },
     {
       "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": 18727.306466607355,
-      "required_pages_per_second_upper_estimate": 0.004332232761712673
+      "required_token_positions_per_second": 18864.736996122636,
+      "required_pages_per_second_upper_estimate": 0.0043640248960213224
     },
     {
       "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": 28090.959699911033,
-      "required_pages_per_second_upper_estimate": 0.001710592063849308
+      "required_token_positions_per_second": 28297.105494183957,
+      "required_pages_per_second_upper_estimate": 0.001723145261157134
     },
     {
       "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": 28090.959699911033,
-      "required_pages_per_second_upper_estimate": 0.0021625198384525104
+      "required_token_positions_per_second": 28297.105494183957,
+      "required_pages_per_second_upper_estimate": 0.002178389512343722
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28090.959699911033,
-      "required_pages_per_second_upper_estimate": 0.3487957050785482
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28297.105494183957,
+      "required_pages_per_second_upper_estimate": 0.3513553459178261
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11735742426563434,
-    "producer_p95_seconds": 0.32239856899832375,
-    "gpu_update_seconds": 0.9842312365030221,
-    "producer_batches_per_second": 8.520977741779129,
-    "gpu_batches_per_second": 1.0160214011831248,
-    "producer_headroom": 8.386612459006,
+    "producer_batch_seconds": 0.18092359124210589,
+    "producer_p95_seconds": 0.25524019799922826,
+    "gpu_update_seconds": 0.9770610639198638,
+    "producer_batches_per_second": 5.527195171921131,
+    "gpu_batches_per_second": 1.0234774845986674,
+    "producer_headroom": 5.400407195169995,
     "sustainable": true,
-    "buffer_seconds": 60.03810542668435,
+    "buffer_seconds": 60.57778596303156,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440615.05307177803,
-    "upload_bytes_per_second": 4962278.859892928,
+    "generation_bytes_per_second": 429934.5762164007,
+    "upload_bytes_per_second": 5008482.897008992,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.54573339571958,
+    "estimated_full_shards_per_hour": 92.25395169133202,
     "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": 1.292083440001079,
-      "bytes": 7124472,
-      "pages": 44,
-      "shards": 2
-    },
     {
       "seconds": 1.2287268149993906,
       "bytes": 6045685,
@@ -562,6 +556,12 @@
       "bytes": 851325,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.4230423680019157,
+      "bytes": 11644215,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -213,3 +213,4 @@
 {"step": 216300, "loss": 0.5820259880274534, "lr": 0.0003, "grad_norm": 0.13927725451570416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31659407184, "tokens_per_second": 148922.5319761339, "data_wait_s": 9.254999895347282e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 216400, "loss": 0.5731642600148916, "lr": 0.0003, "grad_norm": 0.14235341976407173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31674067785, "tokens_per_second": 149381.82045872966, "data_wait_s": 9.686998964753002e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 216500, "loss": 0.5662705302238464, "lr": 0.0003, "grad_norm": 0.16423504628066296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31688855500, "tokens_per_second": 149772.7716992007, "data_wait_s": 1.089499710360542e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 216600, "loss": 0.601808425039053, "lr": 0.0003, "grad_norm": 0.1728315363195944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31703583637, "tokens_per_second": 149834.48276005237, "data_wait_s": 1.4024997653905302e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebe1ba8ac7b946717bc24c8a8a60b4f16da95351eef31c28308d21382b57a488
-size 458667
+oid sha256:7d3d0cd58cea8a16afee080b86a8a4d2099a082dc36163596c2c6686d3e2ec66
+size 460785