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": 77125.04475688896,
-      "required_pages_per_second_upper_estimate": 0.0053356895510405684
+      "required_token_positions_per_second": 77414.44096480278,
+      "required_pages_per_second_upper_estimate": 0.005355710652195762
     },
     {
       "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": 9640.63059461112,
-      "required_pages_per_second_upper_estimate": 0.004036330710704084
+      "required_token_positions_per_second": 9676.805120600347,
+      "required_pages_per_second_upper_estimate": 0.00405147622932577
     },
     {
       "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": 19281.26118922224,
-      "required_pages_per_second_upper_estimate": 0.004015778224663563
+      "required_token_positions_per_second": 19353.610241200695,
+      "required_pages_per_second_upper_estimate": 0.004030846624217869
     },
     {
       "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": 28921.891783833362,
-      "required_pages_per_second_upper_estimate": 0.0017338232512690709
+      "required_token_positions_per_second": 29030.415361801042,
+      "required_pages_per_second_upper_estimate": 0.0017403290740623245
     },
     {
       "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": 28921.891783833362,
-      "required_pages_per_second_upper_estimate": 0.002248549872235472
+      "required_token_positions_per_second": 29030.415361801042,
+      "required_pages_per_second_upper_estimate": 0.0022569871030776872
     },
     {
       "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": 28921.891783833362,
-      "required_pages_per_second_upper_estimate": 0.37097897389506757
+      "required_token_positions_per_second": 29030.415361801042,
+      "required_pages_per_second_upper_estimate": 0.37237099782969746
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.5621277738358401,
-    "producer_p95_seconds": 0.3314579859979858,
-    "gpu_update_seconds": 0.9559540643691421,
-    "producer_batches_per_second": 1.7789549752651663,
-    "gpu_batches_per_second": 1.0460753683388804,
-    "producer_headroom": 1.7005992389344424,
+    "producer_batch_seconds": 0.10431022377346721,
+    "producer_p95_seconds": 0.3185993330007477,
+    "gpu_update_seconds": 0.9523804484168676,
+    "producer_batches_per_second": 9.586787985152077,
+    "gpu_batches_per_second": 1.0500005556206975,
+    "producer_headroom": 9.130269440176574,
     "sustainable": true,
-    "buffer_seconds": 60.22510605525595,
+    "buffer_seconds": 60.95234869867953,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263344.4526432522,
-    "upload_bytes_per_second": 5504956.273912091,
+    "generation_bytes_per_second": 263827.75661417237,
+    "upload_bytes_per_second": 5552431.474383508,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.50758919213462,
+    "estimated_full_shards_per_hour": 56.61129497355345,
     "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": 1.0596985519987356,
-      "bytes": 6289450,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1195497589978913,
       "bytes": 6194218,
@@ -562,6 +556,12 @@
       "bytes": 6334013,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4788876900020114,
+      "bytes": 12280922,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 472700, "loss": 0.5891632437705994, "lr": 0.0003, "grad_norm": 0.1480681987207423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69188784306, "tokens_per_second": 152529.02592809053, "data_wait_s": 1.207699824590236e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 472800, "loss": 0.5954067707061768, "lr": 0.0003, "grad_norm": 0.16198262846134612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69203464134, "tokens_per_second": 152406.81980625115, "data_wait_s": 1.3474000297719613e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 472900, "loss": 0.5664942860603333, "lr": 0.0003, "grad_norm": 0.16556000986740438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69218194319, "tokens_per_second": 153073.5212574671, "data_wait_s": 1.1709998943842947e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 473000, "loss": 0.6076064705848694, "lr": 0.0003, "grad_norm": 0.16217944847790944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69232935689, "tokens_per_second": 153637.8361666627, "data_wait_s": 1.4773999282624573e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d17dea748ac9c84ebb24e1d29aeac14b6bb1658d80a2e259d7a10b63cd46dfab
-size 697449
+oid sha256:a6d69f120bee1b54af103884268550cb9c3b3111215b490f99af6e91905e21b3
+size 699567