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": 77178.08509757729,
-      "required_pages_per_second_upper_estimate": 0.005339359004879933
+      "required_token_positions_per_second": 77105.19068267026,
+      "required_pages_per_second_upper_estimate": 0.00533431600011832
     },
     {
       "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": 9647.260637197161,
-      "required_pages_per_second_upper_estimate": 0.004039106571083827
+      "required_token_positions_per_second": 9638.148835333783,
+      "required_pages_per_second_upper_estimate": 0.00403529164991969
     },
     {
       "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": 19294.521274394323,
-      "required_pages_per_second_upper_estimate": 0.0040185399507129465
+      "required_token_positions_per_second": 19276.297670667565,
+      "required_pages_per_second_upper_estimate": 0.004014744454645411
     },
     {
       "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": 28941.781911591483,
-      "required_pages_per_second_upper_estimate": 0.0017350156340577036
+      "required_token_positions_per_second": 28914.446506001346,
+      "required_pages_per_second_upper_estimate": 0.0017333769182313226
     },
     {
       "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": 28941.781911591483,
-      "required_pages_per_second_upper_estimate": 0.00225009624218124
+      "required_token_positions_per_second": 28914.446506001346,
+      "required_pages_per_second_upper_estimate": 0.002247971034632349
     },
     {
       "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": 28941.781911591483,
-      "required_pages_per_second_upper_estimate": 0.37123410309759347
+      "required_token_positions_per_second": 28914.446506001346,
+      "required_pages_per_second_upper_estimate": 0.37088347386515497
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11484769008546891,
-    "producer_p95_seconds": 0.323269722997793,
-    "gpu_update_seconds": 0.9552970886331877,
-    "producer_batches_per_second": 8.707184265141132,
-    "gpu_batches_per_second": 1.046794774001428,
-    "producer_headroom": 8.317947778682024,
+    "producer_batch_seconds": 0.34737498214070683,
+    "producer_p95_seconds": 0.31961199599754764,
+    "gpu_update_seconds": 0.9562002161881782,
+    "producer_batches_per_second": 2.8787335053246363,
+    "gpu_batches_per_second": 1.0458060802228495,
+    "producer_headroom": 2.7526456001395694,
     "sustainable": true,
-    "buffer_seconds": 60.18371658389083,
+    "buffer_seconds": 60.240613619855225,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322593.2158765324,
-    "upload_bytes_per_second": 4110176.0308090434,
+    "generation_bytes_per_second": 312315.66736456286,
+    "upload_bytes_per_second": 4061690.038520772,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.22099454137782,
+    "estimated_full_shards_per_hour": 67.01567187979379,
     "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": 1.3281459399986488,
-      "bytes": 11209095,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.3028533390024677,
       "bytes": 6867913,
@@ -562,6 +556,12 @@
       "bytes": 7983801,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3452542180020828,
+      "bytes": 6178324,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -344,3 +344,4 @@
 {"step": 876900, "loss": 0.572022557258606, "lr": 3e-05, "grad_norm": 0.3000676411673048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128353810830, "tokens_per_second": 153432.34944826318, "data_wait_s": 1.6711004718672484e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 877000, "loss": 0.5473411083221436, "lr": 3e-05, "grad_norm": 0.30369529390738514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128368665910, "tokens_per_second": 154402.80171194155, "data_wait_s": 1.091499871108681e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 877100, "loss": 0.7569558620452881, "lr": 3e-05, "grad_norm": 0.7409805443240339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128383211391, "tokens_per_second": 150776.6346752798, "data_wait_s": 1.1326999810989946e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 877200, "loss": 0.5050011873245239, "lr": 3e-05, "grad_norm": 0.3247661351031221, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128397863950, "tokens_per_second": 152102.0454064988, "data_wait_s": 7.590999302919954e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1dc8ff371f0edfa2e89a34a9f74094a4a59f2d1a1d6b42a4917517ad33a7bdb
-size 738474
+oid sha256:a2a1a123384cd30ef4343af5aa7b3908847ec2c715a282aab7467dc2c00aee65
+size 740592