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": 76362.74101200452,
-      "required_pages_per_second_upper_estimate": 0.0066189288513228825
+      "required_token_positions_per_second": 76380.03890098877,
+      "required_pages_per_second_upper_estimate": 0.006620428188498938
     },
     {
       "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": 9545.342626500566,
-      "required_pages_per_second_upper_estimate": 0.0038077798893013264
+      "required_token_positions_per_second": 9547.504862623597,
+      "required_pages_per_second_upper_estimate": 0.0038086424376191147
     },
     {
       "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": 19090.68525300113,
-      "required_pages_per_second_upper_estimate": 0.004416294048696657
+      "required_token_positions_per_second": 19095.009725247193,
+      "required_pages_per_second_upper_estimate": 0.004417294439242671
     },
     {
       "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": 28636.0278795017,
-      "required_pages_per_second_upper_estimate": 0.0017437838562346546
+      "required_token_positions_per_second": 28642.51458787079,
+      "required_pages_per_second_upper_estimate": 0.0017441788627412037
     },
     {
       "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": 28636.0278795017,
-      "required_pages_per_second_upper_estimate": 0.0022044806957626914
+      "required_token_positions_per_second": 28642.51458787079,
+      "required_pages_per_second_upper_estimate": 0.002204980060529302
     },
     {
       "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": 28636.0278795017,
-      "required_pages_per_second_upper_estimate": 0.35556362764321614
+      "required_token_positions_per_second": 28642.51458787079,
+      "required_pages_per_second_upper_estimate": 0.35564417085154393
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1440824725859784,
-    "producer_p95_seconds": 0.37487139199947705,
-    "gpu_update_seconds": 0.9654970345866666,
-    "producer_batches_per_second": 6.940469455112034,
-    "gpu_batches_per_second": 1.0357359620768845,
-    "producer_headroom": 6.701002677550006,
+    "producer_batch_seconds": 0.0925444255938146,
+    "producer_p95_seconds": 0.3532037050026702,
+    "gpu_update_seconds": 0.9652783771892731,
+    "producer_batches_per_second": 10.805621122865743,
+    "gpu_batches_per_second": 1.0359705797117618,
+    "producer_headroom": 10.430432422001974,
     "sustainable": true,
-    "buffer_seconds": 60.82631317895999,
+    "buffer_seconds": 60.81253776292421,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420511.52754030196,
-    "upload_bytes_per_second": 5113633.1168090375,
+    "generation_bytes_per_second": 420639.44199251005,
+    "upload_bytes_per_second": 5058810.143163972,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.23198480278772,
+    "estimated_full_shards_per_hour": 90.25943226653553,
     "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": 0.9415042880027613,
-      "bytes": 5881567,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8555704430000333,
       "bytes": 1226801,
@@ -562,6 +556,12 @@
       "bytes": 12207803,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 0.8468366760062054,
+      "bytes": 871307,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -367,3 +367,4 @@
 {"step": 191700, "loss": 0.5890930593013763, "lr": 0.0003, "grad_norm": 0.1589695955741389, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28057264346, "tokens_per_second": 151062.45055904746, "data_wait_s": 1.0711002687457949e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 191800, "loss": 0.5835154447704554, "lr": 0.0003, "grad_norm": 0.1517761281663549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28071917953, "tokens_per_second": 150295.6748934891, "data_wait_s": 8.285001968033612e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 191900, "loss": 0.6047444399446249, "lr": 0.0003, "grad_norm": 0.1811759025553319, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28086780349, "tokens_per_second": 152583.00853259044, "data_wait_s": 1.09329994302243e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 192000, "loss": 0.5808647833764553, "lr": 0.0003, "grad_norm": 0.13895726798050276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28101456290, "tokens_per_second": 149713.7505768038, "data_wait_s": 8.994000381790102e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:482def35fdc64d0fae6f1fbed05e09d8ca6340003d5d92735725549e8b6e7b58
-size 787188
+oid sha256:d2d81b4af019690977202dedf276f462c321e57154c9771183ce732321b579b9
+size 789306