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": 76297.59507963236,
-      "required_pages_per_second_upper_estimate": 0.005278444662420188
+      "required_token_positions_per_second": 77480.50345177944,
+      "required_pages_per_second_upper_estimate": 0.0053602810083308965
     },
     {
       "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": 9537.199384954045,
-      "required_pages_per_second_upper_estimate": 0.0039930262231099335
+      "required_token_positions_per_second": 9685.06293147243,
+      "required_pages_per_second_upper_estimate": 0.004054933602295211
     },
     {
       "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": 19074.39876990809,
-      "required_pages_per_second_upper_estimate": 0.003972694238049278
+      "required_token_positions_per_second": 19370.12586294486,
+      "required_pages_per_second_upper_estimate": 0.004034286392680944
     },
     {
       "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": 28611.598154862135,
-      "required_pages_per_second_upper_estimate": 0.0017152216220031838
+      "required_token_positions_per_second": 29055.188794417292,
+      "required_pages_per_second_upper_estimate": 0.0017418142035208273
     },
     {
       "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": 28611.598154862135,
-      "required_pages_per_second_upper_estimate": 0.0022244259074203927
+      "required_token_positions_per_second": 29055.188794417292,
+      "required_pages_per_second_upper_estimate": 0.0022589131284967865
     },
     {
       "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": 28611.598154862135,
-      "required_pages_per_second_upper_estimate": 0.36699886038996593
+      "required_token_positions_per_second": 29055.188794417292,
+      "required_pages_per_second_upper_estimate": 0.37268876482365915
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 704566,
-    "producer_batch_seconds": 0.13252937544521615,
-    "producer_p95_seconds": 0.25005536200114875,
-    "gpu_update_seconds": 0.9663214144960867,
-    "producer_batches_per_second": 7.545496963526938,
-    "gpu_batches_per_second": 1.0348523638187983,
-    "producer_headroom": 7.291375298871277,
+    "producer_batch_seconds": 0.08263974495304183,
+    "producer_p95_seconds": 0.21377930699964054,
+    "gpu_update_seconds": 0.9515684167680345,
+    "producer_batches_per_second": 12.10071498367072,
+    "gpu_batches_per_second": 1.0508965854462273,
+    "producer_headroom": 11.51465819877278,
     "sustainable": true,
-    "buffer_seconds": 60.87824911325346,
+    "buffer_seconds": 60.90037867315421,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321426.0121983039,
-    "upload_bytes_per_second": 5933281.998072939,
+    "generation_bytes_per_second": 351931.93575088173,
+    "upload_bytes_per_second": 5879710.193253818,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.970539803141,
+    "estimated_full_shards_per_hour": 75.51640085596885,
     "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": 1.1922360469998239,
-      "bytes": 10997251,
-      "pages": 39,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0726386229998752,
-      "bytes": 2419938,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.240178467000078,
       "bytes": 6103593,
@@ -562,6 +550,18 @@
       "bytes": 11430108,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.1905037959986657,
+      "bytes": 6976659,
+      "pages": 37,
+      "shards": 2
+    },
+    {
+      "seconds": 1.609685786999762,
+      "bytes": 5649551,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 499800, "loss": 0.5949404239654541, "lr": 0.0003, "grad_norm": 0.16278332905207246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73153770130, "tokens_per_second": 151713.50425051744, "data_wait_s": 1.686999894445762e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 499900, "loss": 0.5922921299934387, "lr": 0.0003, "grad_norm": 0.1657469942759339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73168359778, "tokens_per_second": 152277.53785663372, "data_wait_s": 1.4310000551631674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 500000, "loss": 0.5788578391075134, "lr": 0.0003, "grad_norm": 0.14864781043951564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73182957636, "tokens_per_second": 151880.7149085412, "data_wait_s": 1.2296000932110474e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 500100, "loss": 0.4749933183193207, "lr": 0.0003, "grad_norm": 0.23643239184333206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73197547318, "tokens_per_second": 140835.37727070332, "data_wait_s": 1.2268003047211096e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8fe631f9cc133c3b611208622db1f396031952371ec93ab2296d7dadbcdf860
-size 320214
+oid sha256:34b536fc77090dc5173fe02c91104bbf8797922084c939e8b669a030a77891c0
+size 322332