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": 74777.02048987121,
-      "required_pages_per_second_upper_estimate": 0.006481482615436286
+      "required_token_positions_per_second": 75390.24677716623,
+      "required_pages_per_second_upper_estimate": 0.006534635515811203
     },
     {
       "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": 9347.127561233901,
-      "required_pages_per_second_upper_estimate": 0.0037287089361871316
+      "required_token_positions_per_second": 9423.78084714578,
+      "required_pages_per_second_upper_estimate": 0.003759287078006135
     },
     {
       "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": 18694.255122467803,
-      "required_pages_per_second_upper_estimate": 0.004324586914929777
+      "required_token_positions_per_second": 18847.56169429156,
+      "required_pages_per_second_upper_estimate": 0.0043600516922176905
     },
     {
       "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": 28041.382683701708,
-      "required_pages_per_second_upper_estimate": 0.0017075730836726617
+      "required_token_positions_per_second": 28271.342541437338,
+      "required_pages_per_second_upper_estimate": 0.0017215764325026316
     },
     {
       "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": 28041.382683701708,
-      "required_pages_per_second_upper_estimate": 0.002158703262506752
+      "required_token_positions_per_second": 28271.342541437338,
+      "required_pages_per_second_upper_estimate": 0.0021764062089248692
     },
     {
       "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": 28041.382683701708,
-      "required_pages_per_second_upper_estimate": 0.348180124460828
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28271.342541437338,
+      "required_pages_per_second_upper_estimate": 0.35103545626776933
     }
   ],
   "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": 0.1450643286401032,
-    "producer_p95_seconds": 0.40732164200016996,
-    "gpu_update_seconds": 0.9859713521212936,
-    "producer_batches_per_second": 6.893493454762033,
-    "gpu_batches_per_second": 1.0142282510019425,
-    "producer_headroom": 6.796787062431009,
+    "producer_batch_seconds": 0.12754903771889303,
+    "producer_p95_seconds": 0.38732158500351943,
+    "gpu_update_seconds": 0.9779514347249797,
+    "producer_batches_per_second": 7.840121869079976,
+    "gpu_batches_per_second": 1.0225456648378668,
+    "producer_headroom": 7.667258430285452,
     "sustainable": true,
-    "buffer_seconds": 60.14425247939891,
+    "buffer_seconds": 60.63298895294874,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326536.84097016894,
-    "upload_bytes_per_second": 5794773.637208084,
+    "generation_bytes_per_second": 326418.5147921825,
+    "upload_bytes_per_second": 5787186.854086453,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.0672046835785,
+    "estimated_full_shards_per_hour": 70.04181464027506,
     "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": 0.9345153110007232,
-      "bytes": 6342336,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2454855190007947,
       "bytes": 7350483,
@@ -562,6 +556,12 @@
       "bytes": 1181916,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.184618612001941,
+      "bytes": 7236384,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -388,3 +388,4 @@
 {"step": 233800, "loss": 0.5587761066854, "lr": 0.0003, "grad_norm": 0.1490171192827732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34221764765, "tokens_per_second": 147533.8362568862, "data_wait_s": 1.1696000001393259e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233900, "loss": 0.5574611648917198, "lr": 0.0003, "grad_norm": 0.14534793742231727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34236369768, "tokens_per_second": 148103.06027596357, "data_wait_s": 1.2780001270584762e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 234000, "loss": 0.6060496270656586, "lr": 0.0003, "grad_norm": 0.13822100157510223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34251105315, "tokens_per_second": 149677.64584513608, "data_wait_s": 1.2583004718180746e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 234100, "loss": 0.6063950918614864, "lr": 0.0003, "grad_norm": 0.15298467270971142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34265917030, "tokens_per_second": 150056.52678457572, "data_wait_s": 1.572600012877956e-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:a95f8a0a551caa6a8e2fc70613b082dd9acc3f22c7393fad90a6129a632d4591
-size 831666
+oid sha256:88768ecd2cf6b14ae5e5ad8c0f229789604bae3d96f7e0ec2d513ef42416c2eb
+size 833784