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": 75712.00783013746,
-      "required_pages_per_second_upper_estimate": 0.006562524948386816
+      "required_token_positions_per_second": 75747.63309942406,
+      "required_pages_per_second_upper_estimate": 0.006565612856437159
     },
     {
       "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": 9464.000978767182,
-      "required_pages_per_second_upper_estimate": 0.003775331489854469
+      "required_token_positions_per_second": 9468.454137428007,
+      "required_pages_per_second_upper_estimate": 0.003777107921424927
     },
     {
       "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": 18928.001957534365,
-      "required_pages_per_second_upper_estimate": 0.004378660131418633
+      "required_token_positions_per_second": 18936.908274856014,
+      "required_pages_per_second_upper_estimate": 0.004380720451185165
     },
     {
       "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": 28392.00293630155,
-      "required_pages_per_second_upper_estimate": 0.0017289240174936955
+      "required_token_positions_per_second": 28405.362412284023,
+      "required_pages_per_second_upper_estimate": 0.0017297375394892747
     },
     {
       "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": 28392.00293630155,
-      "required_pages_per_second_upper_estimate": 0.002185694980130867
+      "required_token_positions_per_second": 28405.362412284023,
+      "required_pages_per_second_upper_estimate": 0.002186723429573394
     },
     {
       "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": 28392.00293630155,
-      "required_pages_per_second_upper_estimate": 0.3525336545476185
+      "required_token_positions_per_second": 28405.362412284023,
+      "required_pages_per_second_upper_estimate": 0.3526995345280309
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07938917989841343,
-    "producer_p95_seconds": 0.2767760950009688,
-    "gpu_update_seconds": 0.9737953346239522,
-    "producer_batches_per_second": 12.596174960864971,
-    "gpu_batches_per_second": 1.0269098284252587,
-    "producer_headroom": 12.266096410997353,
+    "producer_batch_seconds": 0.06300009664832373,
+    "producer_p95_seconds": 0.208494583999709,
+    "gpu_update_seconds": 0.9733373437982789,
+    "producer_batches_per_second": 15.8729915222536,
+    "gpu_batches_per_second": 1.027393027064671,
+    "producer_headroom": 15.449775406402917,
     "sustainable": true,
-    "buffer_seconds": 60.37531074668504,
+    "buffer_seconds": 60.34691531549329,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 535424.1222911193,
+    "generation_bytes_per_second": 548852.016358536,
     "upload_bytes_per_second": 5944322.7863628855,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.88955260801491,
+    "estimated_full_shards_per_hour": 117.77086608950671,
     "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": [

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -120,3 +120,4 @@
 {"step": 97000, "loss": 0.6435183398425579, "lr": 0.0003, "grad_norm": 0.28237957908045064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14194545126, "tokens_per_second": 148950.47540112672, "data_wait_s": 1.564100239193067e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 97100, "loss": 0.5977706927806139, "lr": 0.0003, "grad_norm": 0.12892321977221902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14209057025, "tokens_per_second": 147808.54666147774, "data_wait_s": 1.7684000340523198e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 97200, "loss": 0.6232582665979862, "lr": 0.0003, "grad_norm": 0.15621962243480964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14223550833, "tokens_per_second": 147672.04303420355, "data_wait_s": 6.507001671707258e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 97300, "loss": 0.6078848466277122, "lr": 0.0003, "grad_norm": 0.145436077212289, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14238146498, "tokens_per_second": 148191.2769510072, "data_wait_s": 1.0234001820208505e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14d5da7036023b8b6ac06d31cb2424062b37aba2792281da229817876f4fa8d0
-size 260127
+oid sha256:d5564c84a8669d6fe4ebfad516828991e01d41806804fbb8ebc0600a3b0cc717
+size 262245