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": 74216.19837367054,
-      "required_pages_per_second_upper_estimate": 0.006432871975794671
+      "required_token_positions_per_second": 72773.65439671188,
+      "required_pages_per_second_upper_estimate": 0.006307835919966178
     },
     {
       "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": 9277.024796708818,
-      "required_pages_per_second_upper_estimate": 0.003700743895288343
+      "required_token_positions_per_second": 9096.706799588985,
+      "required_pages_per_second_upper_estimate": 0.003628812350242933
     },
     {
       "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": 18554.049593417636,
-      "required_pages_per_second_upper_estimate": 0.004292152833317056
+      "required_token_positions_per_second": 18193.41359917797,
+      "required_pages_per_second_upper_estimate": 0.004208726043026432
     },
     {
       "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": 27831.074390126454,
-      "required_pages_per_second_upper_estimate": 0.0016947664119962697
+      "required_token_positions_per_second": 27290.120398766954,
+      "required_pages_per_second_upper_estimate": 0.0016618251520887238
     },
     {
       "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": 27831.074390126454,
-      "required_pages_per_second_upper_estimate": 0.0021425131478967118
+      "required_token_positions_per_second": 27290.120398766954,
+      "required_pages_per_second_upper_estimate": 0.0021008690121853677
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27831.074390126454,
-      "required_pages_per_second_upper_estimate": 0.3455687993112042
+      "required_token_positions_per_second": 27290.120398766954,
+      "required_pages_per_second_upper_estimate": 0.33885196119506505
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10354812821091741,
-    "producer_p95_seconds": 0.20452078900052584,
-    "gpu_update_seconds": 0.9934219431287423,
-    "producer_batches_per_second": 9.657345017025298,
-    "gpu_batches_per_second": 1.0066216142262172,
-    "producer_headroom": 9.593818452277949,
+    "producer_batch_seconds": 0.46887441861719026,
+    "producer_p95_seconds": 0.20191862800129456,
+    "gpu_update_seconds": 1.0131138886895206,
+    "producer_batches_per_second": 2.1327672406381466,
+    "gpu_batches_per_second": 0.9870558593304021,
+    "producer_headroom": 2.160736112832531,
     "sustainable": true,
-    "buffer_seconds": 60.59873853085328,
+    "buffer_seconds": 60.78683332137123,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276599.71288855816,
-    "upload_bytes_per_second": 5762539.361239716,
+    "generation_bytes_per_second": 255113.53634045398,
+    "upload_bytes_per_second": 5784141.317043891,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.351859474111166,
+    "estimated_full_shards_per_hour": 54.74142615947928,
     "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": 1.4607127819999732,
-      "bytes": 6008702,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.7417863239998042,
       "bytes": 879449,
@@ -562,6 +556,12 @@
       "bytes": 7398482,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2963822889996663,
+      "bytes": 6635646,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 367600, "loss": 0.6230793055146933, "lr": 0.0003, "grad_norm": 0.15854659041598212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53806626944, "tokens_per_second": 149560.88910709284, "data_wait_s": 8.148999768309295e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 367700, "loss": 0.6268834322690964, "lr": 0.0003, "grad_norm": 0.15775056401275156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53821386282, "tokens_per_second": 150685.15193859203, "data_wait_s": 1.1766000170609914e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 367800, "loss": 0.5165554955601692, "lr": 0.0003, "grad_norm": 0.1576865568884214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53836039685, "tokens_per_second": 125656.96931929927, "data_wait_s": 9.327999578090385e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 367900, "loss": 0.5798914544284344, "lr": 0.0003, "grad_norm": 0.17950501109424952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53850531284, "tokens_per_second": 146687.46292830593, "data_wait_s": 1.8110000382876024e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6d2dcac8f2b255fa527224e6ee6a0760d000e9c4b5e5cbbb0328dec28e48dc9
-size 272835
+oid sha256:97ca1247a6adf32a72d77b57e240c998c52eadfaf9c4620dff74729f7b805a5b
+size 274953